The BGK Relaxation-Time Approximation

The Boltzmann collision integral is a quadratic, five-fold integral over scattering angles and partner velocities — analytically hopeless except in special limits. In 1954 Bhatnagar, Gross and Krook proposed a radical simplification: throw the integral away and replace it with a single relaxation term $-(f-f^{\mathrm{eq}})/\tau$ that pulls the distribution toward a local Maxwellian at one rate $1/\tau$ . The astonishing thing is how much survives. Number, momentum and energy are still conserved, the H-theorem still holds, sound still propagates with the right speed, and a Chapman–Enskog expansion still delivers the Navier–Stokes equations with explicit transport coefficients. The one non-negotiable design rule is that the target $f^{\mathrm{eq}}$ must be a local Maxwellian built from $f$ 's own instantaneous density $n$ , drift $\mathbf u$ , and temperature $T$ . This self-consistency is exactly what protects the collisional invariants: because $f^{\mathrm{eq}}$ shares $f$ 's first five velocity moments, the relaxation term integrates to zero against $1,\mathbf v,v^2$ . You cannot relax toward something that has different mass or energy than you have right now. Everything else — the stresses and heat fluxes, the non-conserved higher moments — decays at rate $1/\tau$ , and hydrodynamics is the slow dynamics of what is left protected. That single relaxation time is also the model's Achilles heel. Because the same $\tau$ that sets the viscosity $\eta=p\tau$ also sets the thermal conductivity $\kappa=\tfrac52(k_B/m)p\tau$ , their ratio — the Prandtl number $\mathrm{Pr}=\eta c_p/\kappa$ — is forced to exactly $1$ . Real monatomic gases have $\mathrm{Pr}\approx 2/3$ , so BGK overpredicts by a clean factor of $3/2$ . This one number is the sharpest diagnostic of the single-relaxation-time assumption, and it motivates the ES-BGK and Shakhov refinements. Conceptually BGK sits beside the master equation and Fokker–Planck as a relaxation picture of the approach to equilibrium, and it is the computational heart of the lattice Boltzmann method.

⚡ Key Takeaways

The Bhatnagar–Gross–Krook (BGK) model replaces the full Boltzmann collision integral by a single relaxation term $-(f-f^{\mathrm{eq}})/\tau$ , driving the distribution toward a local Maxwellian at rate $1/\tau$ .
The target $f^{\mathrm{eq}}$ is built from the instantaneous local density, drift, and temperature — this self-consistency is what conserves mass, momentum, and energy exactly.
A Chapman–Enskog expansion yields Navier–Stokes with $\eta=p\tau$ and $\kappa=\tfrac52(k_B/m)p\tau$ ; a single $\tau$ ties them together.
The model's signature defect: one relaxation time forces the Prandtl number to exactly $\mathrm{Pr}=1$ , whereas monatomic gases have $\mathrm{Pr}\approx 2/3$ — a factor $3/2$ error.
BGK keeps the H-theorem, linearizes cleanly, and is the workhorse of the lattice Boltzmann method and solid-state transport (the relaxation-time approximation).
Key idea: summarize all collisions by one number, a relaxation time $\tau$ , that pulls $f$ toward the local Maxwellian built from $f$ 's own conserved moments; conservation and the H-theorem survive, hydrodynamics emerges with transport coefficients $\propto\tau$ , but the fixed $\mathrm{Pr}=1$ is the price.

💡 The Core Insight

Summarize all collisions by one number — a relaxation time $\tau$ — that pulls $f$ toward the local Maxwellian $f^{\mathrm{eq}}$ built from $f$ 's own conserved moments. Conservation and the H-theorem survive; hydrodynamics emerges with transport coefficients $\propto\tau$ ; the single $\tau$ forces the Prandtl number to exactly $1$ .

The physical picture

Picture a gas slightly out of equilibrium. Whatever the precise scattering mechanism, the net effect of many collisions is to scramble correlations and push the velocity distribution back toward a Maxwellian centered on the local flow. BGK says: don't track the scrambling in detail — just assume the distribution decays exponentially toward that local Maxwellian with one time constant $\tau$ . It is the kinetic-theory analog of Newton's law of cooling for the whole distribution function. The subtlety is that the target moves. As the gas flows and heats, the local Maxwellian's parameters $n(\mathbf r,t)$ , $\mathbf u(\mathbf r,t)$ , $T(\mathbf r,t)$ change. BGK insists the target always matches $f$ 's own instantaneous density, momentum, and energy. Think of it as five protected directions in distribution space — spanned by $1,v_x,v_y,v_z,v^2$ — that the relaxation cannot touch, plus all remaining directions (stresses, heat flux, higher moments) that decay at rate $1/\tau$ . Hydrodynamics is the slow dynamics of the protected modes; transport coefficients measure how fast the non-protected modes leak into them.

Figure

Relaxation of a perturbed distribution toward the local Maxwellian, with the fixed Pr=1 defect as the outlier

The same $\tau$ that sets viscosity sets thermal conductivity, because every non-conserved moment relaxes at one rate. Real collisions relax momentum and energy at slightly different effective rates, which is why $\mathrm{Pr}=1$ in BGK but $\approx 2/3$ in argon — a clean diagnostic of the single-relaxation-time assumption. The figure's inset makes the defect visible: the red BGK bar sits a factor $3/2$ above the true monatomic value.

Figure

The five protected moment directions versus the fast-decaying stress and heat-flux channels

Why does the arrow of relaxation point toward $f^{\mathrm{eq}}$ rather than away from it? Because the relaxation term satisfies an H-theorem: the operator $-(f-f^{\mathrm{eq}})/\tau$ can only decrease Boltzmann's $H=\int f\ln f\,\mathrm d^3v$ , and the minimum is exactly the local Maxwellian. The single time constant thus carries genuine physics — it encodes the mean free time between collisions, $\tau\sim\ell/v_{\mathrm{th}}$ , and through it the whole slow structure of transport.

The core equations

The Boltzmann equation splits into free streaming (left) and collisions (right); BGK's entire content is the replacement of the right-hand side. The streaming operator is untouched — particles still drift and accelerate under forces — but every scattering event is summarized by exponential pull toward the local Maxwellian.

\frac{\partial f}{\partial t}+\mathbf v\cdot\nabla_{\mathbf r} f+\frac{\mathbf F}{m}\cdot\nabla_{\mathbf v} f=-\frac{f-f^{\mathrm{eq}}}{\tau} \qquad (1)

where: $f(\mathbf r,\mathbf v,t)$ — one-particle phase-space distribution; $\mathbf F$ — external force per particle; $m$ — particle mass; $\tau$ — single collision relaxation time; $f^{\mathrm{eq}}$ — local Maxwellian built from the moments of $f$ . The right side reads: relax toward local equilibrium at rate $1/\tau$ , and nothing more. The target is not a fixed Maxwellian but a moving one whose five parameters are recomputed from $f$ at every point and instant. This is the self-consistency that makes conservation automatic.

f^{\mathrm{eq}}(\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 (2)

where: $n=\int f\,\mathrm d^3v$ — local number density; $\mathbf u=\frac1n\int\mathbf v f\,\mathrm d^3v$ — local mean velocity; $T$ from $\frac32 nk_BT=\frac12 m\!\int(\mathbf v-\mathbf u)^2 f\,\mathrm d^3v$ — local temperature; $k_B$ — Boltzmann constant. Because $f^{\mathrm{eq}}$ inherits $f$ 's own moments, it agrees with $f$ on exactly the conserved quantities. The conservation statement is that the collision term integrates to zero against each collisional invariant $\chi\in\{1,\mathbf v,v^2\}$ — mass, momentum, and energy are untouched by collisions.

\int \chi(\mathbf v)\,(f-f^{\mathrm{eq}})\,\mathrm d^3v=0,\qquad \chi\in\{1,\mathbf v,v^2\} \qquad (3)

where: $\chi$ — a collisional invariant. This is not an assumption imposed afterward; it holds by construction because $f^{\mathrm{eq}}$ was built to share those moments. The Chapman–Enskog expansion in the Knudsen number then extracts hydrodynamics. Its leading correction gives the shear viscosity as pressure times the relaxation time — momentum diffuses at the rate collisions allow.

\eta=n k_B T\,\tau=p\,\tau \qquad (4)

where: $\eta$ — dynamic shear viscosity; $p=nk_BT$ — local pressure. Reading the same machinery on the heat-flux moment gives the thermal conductivity.

\kappa=\frac{5}{2}\frac{n k_B^2 T}{m}\,\tau=\frac{5}{2}\frac{k_B}{m}\,p\,\tau \qquad (5)

where: $\kappa$ — thermal conductivity. Both $\eta$ and $\kappa$ are linear in the same $\tau$ , and that shared timescale forces their ratio.

\mathrm{Pr}=\frac{\eta\,c_p}{\kappa}=\frac{(p\tau)\big(\tfrac52 k_B/m\big)}{\tfrac52(k_B/m)\,p\tau}=1 \qquad (6)

where: $\mathrm{Pr}$ — Prandtl number (ratio of momentum to thermal diffusivity); $c_p=\tfrac52 k_B/m$ — monatomic specific heat at constant pressure per unit mass. Everything cancels: BGK predicts exactly $1$ , while argon measures $\approx 2/3$ . That factor $3/2$ discrepancy is the model's defining limitation.

Derivation AChapman-Enskog expansion: from BGK to Navier-Stokes viscosity

The goal is to show that the crude single-time relaxation, expanded to first order in the Knudsen number, reproduces the Navier–Stokes momentum flux with the explicit coefficient $\eta=p\tau$ . We write $f=f^{\mathrm{eq}}(1+\phi)$ with $\phi$ of order the Knudsen number $\mathrm{Kn}=\ell/L$ , treat the streaming operator as the perturbation that must be balanced by relaxation, solve for the deviation $\phi$ , and close the momentum-flux moment integral over $f^{\mathrm{eq}}$ . The Gaussian moments do all the work; the coefficient falls out with no free parameters.

Step 1. At leading order the streaming of the local Maxwellian must be balanced by the relaxation of the deviation: $(\partial_t+\mathbf v\cdot\nabla)f^{\mathrm{eq}}\approx -f^{\mathrm{eq}}\phi/\tau$ . Why: to lowest order in $\mathrm{Kn}$ the left side is small, so $\phi$ must itself be of order $\mathrm{Kn}$ for the balance to hold.

Step 2. Solve for the deviation: $\phi=-\tau\,\frac{1}{f^{\mathrm{eq}}}(\partial_t+\mathbf v\cdot\nabla)f^{\mathrm{eq}}$ , and eliminate the explicit time derivative using the zeroth-order (Euler) equations of motion. Why: on the slow hydrodynamic manifold the time dependence of $n,\mathbf u,T$ is slaved to their spatial gradients through the conservation laws.

Step 3. Extract the traceless symmetric rate of strain: $\phi\propto -\frac{m\tau}{k_BT}\,\bar c_i\bar c_j\,\partial_i u_j$ with $\bar{\mathbf c}=\mathbf v-\mathbf u$ . Why: only the shear part of $\partial_i u_j$ couples to the traceless second moment; the bulk part feeds pressure, not shear stress.

Step 4. Compute the momentum flux $\Pi_{ij}=\int m\bar c_i\bar c_j\,f^{\mathrm{eq}}\phi\,\mathrm d^3v=-\eta(\partial_i u_j+\partial_j u_i-\tfrac23\delta_{ij}\nabla\!\cdot\!\mathbf u)$ . Why: the fourth-order Gaussian moment $\langle\bar c_i\bar c_j\bar c_k\bar c_l\rangle$ over $f^{\mathrm{eq}}$ contributes factors of $(k_BT/m)^2$ , and Wick pairing assembles the Navier–Stokes tensor.

Step 5. Read off $\eta=nk_BT\,\tau=p\,\tau$ ; the same closure on the heat-flux moment gives $\kappa=\tfrac52(k_B/m)p\tau$ . Why: the moment integrals fix the numerical coefficients uniquely once $\tau$ and the Maxwellian are specified. *Verified: the coefficient $\eta=p\tau$ and $\kappa=\tfrac52(k_B/m)p\tau$ match Chapman & Cowling and Kardar Ch. 3, and are consistent with the elementary estimate $\eta\sim nm v_{\mathrm{th}}\ell$ with $\ell=v_{\mathrm{th}}\tau$ ; symbolic checks in verify_aq-sm-bgk-relaxation-approximation.py.*

Derivation BSpatially homogeneous relaxation: exact exponential decay and the H-theorem

The goal is to solve the BGK equation exactly in the one regime where it is analytically closed — a spatially uniform, force-free gas — and to show that the same relaxation term drives Boltzmann's $H$ monotonically downhill. With $\nabla f=0$ and $\mathbf F=0$ , the conserved moments $n,\mathbf u,T$ are constants of motion, so $f^{\mathrm{eq}}$ is time-independent and the equation for each velocity becomes an independent, decoupled linear first-order ODE. We then insert the solution into $\dot H$ and use a pointwise convexity inequality.

Step 1. With $f^{\mathrm{eq}}$ fixed, the equation reduces to $\partial_t(f-f^{\mathrm{eq}})=-(f-f^{\mathrm{eq}})/\tau$ . Why: streaming and forcing both vanish, and the target no longer moves because its parameters are conserved.

Step 2. Integrate mode by mode: $f(\mathbf v,t)=f^{\mathrm{eq}}(\mathbf v)+[f(\mathbf v,0)-f^{\mathrm{eq}}(\mathbf v)]\,e^{-t/\tau}$ . Why: each velocity $\mathbf v$ is a decoupled scalar relaxing exponentially — the kinetic analog of Ornstein–Uhlenbeck relaxation of a single mode.

Step 3. Confirm moment consistency: $\int\chi f(\mathbf v,t)\,\mathrm d^3v=\int\chi f^{\mathrm{eq}}\,\mathrm d^3v$ for all $t$ , because $\int\chi(f(0)-f^{\mathrm{eq}})\,\mathrm d^3v=0$ . Why: the exponentially decaying part carries no mass, momentum, or energy, so the conserved quantities stay pinned.

Step 4. For the H-theorem, insert the relaxation term into $\dot H=\int(\ln f+1)(\partial_t f)\,\mathrm d^3v$ to get $\dot H=-\frac1\tau\int(\ln f-\ln f^{\mathrm{eq}})(f-f^{\mathrm{eq}})\,\mathrm d^3v$ . Why: the $\ln f^{\mathrm{eq}}+1$ piece integrates to zero because $\ln f^{\mathrm{eq}}$ is a linear combination of the collisional invariants, against which $f-f^{\mathrm{eq}}$ has vanishing moments.

Step 5. Apply $(\ln a-\ln b)(a-b)\ge 0$ pointwise with $a=f$ , $b=f^{\mathrm{eq}}$ : the integrand is non-negative, so $\dot H\le 0$ , with equality iff $f=f^{\mathrm{eq}}$ . Why: the logarithm is monotone, so the product of the sign-aligned factors is never negative — collisions can only lower $H$ . *Verified: at $t=0$ the solution returns the initial data and at $t\to\infty$ the local Maxwellian; the H-theorem sign follows from $(\ln a-\ln b)(a-b)\ge0$ sampled numerically; verify_aq-sm-bgk-relaxation-approximation.py.*

Limiting cases

With $\mathrm{Kn}=\ell/L$ the Knudsen number, $\ell\sim v_{\mathrm{th}}\tau$ the mean free path, and $L$ the macroscopic scale:

Case	Result	Meaning
Hydrodynamic limit $\mathrm{Kn}\to 0$	$f\to f^{\mathrm{eq}}(1+\mathrm{Kn}\,f^{(1)}+\dots)$	Frequent collisions: $f$ is everywhere near local Maxwellian; Chapman–Enskog gives Euler (order 0) then Navier–Stokes (order 1).
Free-molecular limit $\mathrm{Kn}\to\infty$	$\partial_t f+\mathbf v\cdot\nabla f\approx 0$	Collisions negligible; particles stream ballistically and the relaxation term is subdominant.
Uniform perturbation	$\delta f(t)=\delta f(0)\,e^{-t/\tau}$	A non-conserved moment decays as a single exponential — the OU-like signature of BGK.
Solid-state (electrons/phonons)	$\sigma=ne^2\tau/m$	The relaxation-time approximation gives the Drude conductivity, the condensed-matter incarnation of BGK.

⚠️ Common pitfall

The local equilibrium $f^{\mathrm{eq}}$ depends on $n,\mathbf u,T$ , which are themselves moments of $f$ , so the BGK operator is nonlinear in $f$ even though it looks linear — you cannot treat $f^{\mathrm{eq}}$ as a fixed target. And a single $\tau$ rigidly sets $\mathrm{Pr}=1$ : do not expect quantitatively correct viscosity and conductivity simultaneously for a real gas. The fix is ES-BGK or Shakhov, which introduce a second parameter to free the Prandtl number.

Worked examples

Example 1 — The Prandtl-number defect in numbers. Compute the BGK Prandtl number and compare with argon.

Step 1. From the transport coefficients, $\eta=p\tau$ and $\kappa=\tfrac52(k_B/m)p\tau$ , and the monatomic $c_p=\tfrac52 k_B/m$ .

Step 2. Form the ratio: $\mathrm{Pr}=\eta c_p/\kappa=(p\tau)(\tfrac52 k_B/m)\big/[\tfrac52(k_B/m)p\tau]$ . Every factor of $p$ , $\tau$ , $k_B$ , $m$ cancels.

Step 3. The result is $\mathrm{Pr}_{\mathrm{BGK}}=\mathbf{1}$ , independent of pressure, temperature, and gas species. Measured argon has $\mathrm{Pr}\approx 0.67$ , so BGK overpredicts by $1/0.67\approx\mathbf{1.49}$ — very nearly the exact factor $3/2$ that $\mathrm{Pr}=1$ versus $\mathrm{Pr}=2/3$ demands. BGK is qualitatively right and quantitatively off by 3/2 in the momentum-to-heat transport ratio.

Example 2 — Viscosity of a dilute gas. Estimate $\eta$ for a gas at $1\,$ atm with collision time $\tau=2.0\times10^{-10}\,$ s.

Step 1. Use $\eta=p\tau$ with $p=1.0\times10^{5}\,$ Pa.

Step 2. Then $\eta=(1.0\times10^{5})(2.0\times10^{-10})=\mathbf{2.0\times10^{-5}\ Pa\cdot s}$ , the right order of magnitude for air or argon (measured $\approx1.8$ – $2.2\times10^{-5}\,$ Pa·s at room temperature).

Step 3. Check the Prandtl consistency with argon mass $m=6.63\times10^{-26}\,$ kg: $\kappa=\tfrac52(k_B/m)p\tau$ and $c_p=\tfrac52 k_B/m$ give $\mathrm{Pr}=\eta c_p/\kappa=\mathbf{1}$ exactly, the mass cancelling as required. The single collision time simultaneously fixes viscosity and conductivity — the source of both the model's economy and its Prandtl defect.

relaxation time approximation

kinetic model, 1954 · The relaxation-time approximation replaces the intractable Boltzmann collision integral by a single term

-(f-f^{\mathrm{eq}})/\tau

that relaxes the distribution toward local equilibrium at one rate. Introduced by Bhatnagar, Gross and Krook in 1954, it is now the default collision model across rarefied gas dynamics, the lattice Boltzmann method, and solid-state transport (where

f^{\mathrm{eq}}

becomes the local Fermi–Dirac or Bose–Einstein distribution). Its enduring appeal is that it preserves conservation laws and the H-theorem while reducing a five-fold integral to one algebraic operation.

⏳ History

1872

— Boltzmann publishes the transport equation and the H-theorem, defining the collision integral the BGK model later approximates

1916

— Chapman and Enskog independently develop the systematic expansion that extracts hydrodynamics and transport coefficients

1954

— Bhatnagar, Gross and Krook propose the single-relaxation-time collision model in Phys. Rev. 94, 511

1966

— Holway proposes the ellipsoidal-statistical (ES-BGK) model, restoring the correct Prandtl number with an anisotropic Gaussian target

1968

— Shakhov adds a heat-flux correction to the target, giving a tunable Prandtl number

1992

— Qian, d'Humières and Lallemand formulate the lattice BGK method (Europhys. Lett. 17, 479), making the BGK operator the workhorse of lattice-Boltzmann computational fluid dynamics

Where this lives today

The BGK operator is the beating heart of the lattice Boltzmann method, now a mainstream computational fluid dynamics tool for porous-media flow, multiphase systems, and blood flow in complex geometries: on a discrete velocity lattice the collision step is simply $f_i\to f_i-\frac{1}{\tilde\tau}(f_i-f_i^{\mathrm{eq}})$ , local and trivially parallel. Its known $\mathrm{Pr}=1$ limitation drives the widespread use of multiple-relaxation-time and ES-BGK variants when heat transport must be quantitative. The same relaxation-time approximation, with $f^{\mathrm{eq}}$ a local Fermi–Dirac distribution, underlies the Boltzmann transport equation for electrons and phonons in solids — it is what turns microscopic scattering into the Drude conductivity $\sigma=ne^2\tau/m$ , thermal conductivity, and Seebeck coefficients that appear in every semiconductor-device and thermoelectric simulation. Modern first-principles transport codes compute $\tau(\mathbf k)$ from electron–phonon coupling and feed it straight into this BGK-shaped equation.

Going further

References. Bhatnagar, Gross & Krook, Phys. Rev. 94, 511 (1954) · Kardar, Statistical Physics of Particles, Ch. 3 (Kinetic Theory) · Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 13–14 · Reif, Fundamentals of Statistical and Thermal Physics, Ch. 12–14 · Chapman & Cowling, The Mathematical Theory of Non-Uniform Gases · Cercignani, The Boltzmann Equation and Its Applications.

Provenance. Equations standard from Bhatnagar–Gross–Krook (1954) and textbook kinetic theory. A verifier should: (1) confirm the BGK transport coefficients $\eta=p\tau$ , $\kappa=\tfrac52(k_B/m)p\tau$ and $\mathrm{Pr}=1$ via Chapman–Enskog (Kardar Ch. 3; Chapman & Cowling); (2) verify the H-theorem inequality using $(\ln a-\ln b)(a-b)\ge0$ ; (3) check the homogeneous exponential solution by direct substitution; (4) note the empirical $\mathrm{Pr}\approx2/3$ for monatomic gases (argon) as the contrast value. No fabricated constants used. This article recalls, by name only, the boltzmann-transport-equation (whose full collision integral BGK replaces) and the h-theorem (which the relaxation term still satisfies); the local-Maxwellian and Chapman–Enskog machinery are referenced as standard kinetic theory.

Prerequisites: the master equation · the Fokker–Planck equation · the Ornstein–Uhlenbeck process. Leads to: the diffusion equation · the fluctuation-dissipation theorem · the Einstein relation.

Is the BGK relaxation time the same as the mean free time between collisions?

Essentially yes, up to order-unity factors. The physical content of $\tau$ is the timescale over which collisions erase non-equilibrium structure, and dimensionally $\tau\sim\ell/v_{\mathrm{th}}$ with $\ell$ the mean free path. The kinematic viscosity then reads $\nu=\eta/(nm)=(k_BT/m)\tau=v_{\mathrm{th}}^2\tau\sim v_{\mathrm{th}}\ell$ , reproducing the elementary kinetic-theory estimate $\eta\sim nm v_{\mathrm{th}}\ell$ . Fitting $\tau$ to one measured transport coefficient (say viscosity) then fixes the model; the $\mathrm{Pr}=1$ defect means the other coefficient will be off by the corresponding factor.

Why does the local Maxwellian have to track f's moments — couldn't we relax to a fixed global equilibrium?

Relaxing to a fixed global Maxwellian would violate conservation whenever the gas is locally hotter, denser, or moving. Collisions in a real gas are local and instantaneous in the kinetic picture: they redistribute velocities at a point without changing the number, momentum, or energy there. Only a target that matches $f$ 's own local $n,\mathbf u,T$ can leave those five quantities untouched. Relax to a global Maxwellian and you would spuriously transport mass and energy through the collision term, breaking the very conservation laws hydrodynamics is built on.

How can a model this crude still give the correct Navier-Stokes equations?

Because hydrodynamics is governed by the slow dynamics of the conserved densities, and those are insensitive to the microscopic details of collisions — they care only that collisions relax non-conserved moments on some timescale $\tau$ . The Chapman–Enskog expansion makes this precise: to first order in the Knudsen number, any collision operator that conserves the invariants and drives relaxation produces Navier–Stokes structure. BGK gets the form of the transport laws exactly right and only the ratio of the coefficients wrong.

What exactly goes wrong to make Pr=1 instead of 2/3?

In a real gas, a velocity-space perturbation carrying shear stress and one carrying heat flux relax at different effective rates — the collision operator has different eigenvalues in those two channels. BGK replaces the whole eigenvalue spectrum with a single number $1/\tau$ , so stress and heat flux relax identically. Since $\eta$ and $\kappa$ are then both proportional to the same $\tau$ , their ratio is frozen at unity. The ES-BGK and Shakhov models restore two independent relaxation rates and recover $\mathrm{Pr}\approx 2/3$ .

Does the BGK model satisfy an H-theorem like the full Boltzmann equation?

Yes, and this is one of its most important features. Inserting the relaxation term into $\dot H$ gives $\dot H=-\frac1\tau\int(\ln f-\ln f^{\mathrm{eq}})(f-f^{\mathrm{eq}})\,\mathrm d^3v\le 0$ , using the elementary inequality $(\ln a-\ln b)(a-b)\ge0$ . Equality holds only at $f=f^{\mathrm{eq}}$ . So BGK, despite discarding the detailed collision integral, still guarantees monotone approach to local equilibrium and thermodynamic consistency — which is why it is a legitimate physical model and not just a numerical convenience.

How does BGK connect to the relaxation-time approximation in solid-state physics?

They are the same idea in different clothing. In a metal or semiconductor, the electron distribution $f(\mathbf k)$ obeys a Boltzmann transport equation whose collision term is often approximated as $-(f-f^0)/\tau$ , with $f^0$ the local Fermi–Dirac distribution. This yields the Drude conductivity $\sigma=ne^2\tau/m$ , the Wiedemann–Franz law, and thermoelectric coefficients. The gas-kinetic $\eta,\kappa\propto\tau$ and the electronic $\sigma\propto\tau$ share the identical structure: transport is proportional to the time between scattering events.

When does the BGK equation fail and you need the full Boltzmann collision integral?

When the details of the collision cross-section matter quantitatively — for the correct Prandtl number, for the precise temperature dependence of transport coefficients, or in strongly non-equilibrium situations (shock interiors, high-Knudsen microflows) where the distribution is far from any local Maxwellian and higher moments cannot be captured by one relaxation rate. In those cases one solves the full Boltzmann equation (e.g. by direct simulation Monte Carlo) or uses multi-relaxation-time refinements. BGK remains excellent for near-equilibrium hydrodynamics and as a computational engine.