boltzmann transport equation

The Boltzmann transport equation evolves a distribution function through phase space, balancing deterministic motion against scattering. For a semiclassical band and carrier species, $\partial_t f+\dot{\mathbf r}\cdot\nabla_{\mathbf r}f+\dot{\mathbf k}\cdot\nabla_{\mathbf k}f=C[f]+S$ says that occupation is advected by real-space velocity and reciprocal-space force while collisions and explicit sources redistribute particles. Densities, currents, stress, energy flow, mobility, diffusion, viscosity, and thermal conductivity are moments of $f$; they are outputs of a declared kinetic model rather than independent empirical fields. ```svg ``` **The distribution function is the central unknown.** In a classical dilute gas, $f(\mathbf r,\mathbf p,t)$ gives expected particles per phase-space volume under a stated normalization. For semiclassical electrons, $f_n(\mathbf r,\mathbf k,t)$ is the occupation probability of a Bloch state in band $n$, bounded between zero and one by Pauli exclusion. Phonon distributions are bosonic and unbounded. Confusing probability, occupation, density of states, and particle density introduces missing phase-space measures or degeneracy factors before any collision physics is considered. **Phase-space streaming is a Liouville derivative along trajectories.** Without collisions or sources, $df/dt=0$ along characteristics generated by $\dot{\mathbf r}$ and $\dot{\mathbf k}$. A packet can move and distort in coordinate projections while its fine-grained phase-space occupation is conserved. This is not the statement that macroscopic density is constant at a fixed point. Boundaries, coarse graining, and collisions produce irreversible-looking relaxation even though microscopic Hamiltonian motion preserves phase-space volume under appropriate assumptions. **Semiclassical band dynamics supplies velocity and force.** For band energy $\varepsilon_n(\mathbf k)$, group velocity is $\mathbf v_n=\hbar^{-1}\nabla_{\mathbf k}\varepsilon_n$. Under electric and magnetic fields, $\hbar\dot{\mathbf k}=q(\mathbf E+\mathbf v\times\mathbf B)$ for charge $q$, with sign stated explicitly. Berry curvature can add anomalous velocity and phase-space corrections. The semiclassical picture requires wave packets localized over scales large compared with a lattice constant and interband transitions weak enough for a band label to remain meaningful. **The collision operator contains the real modeling burden.** $C[f]$ represents transitions caused by phonons, impurities, defects, carrier–carrier interactions, boundaries, chemistry, or radiation. It may be linear for scattering from a fixed bath or nonlinear when populations scatter from one another. A credible operator declares transition rates, state counting, energy and momentum selection, Pauli blocking or Bose stimulation, and what the environment absorbs. Writing “collision term” without these choices does not close the BTE. **Moments translate kinetic detail into measurable fields.** Integrating $f$ over momentum or wave vector gives number density; weighting by charge times velocity gives electrical current; weighting by energy relative to electrochemical potential gives heat current; weighting by momentum flux gives stress. Normalization includes band, spin, valley, and Brillouin-zone factors. Moments discard information, so many distinct distributions share the same density and mean velocity. A closure based only on a few moments is justified only when neglected angular and energy structure relaxes rapidly. **The BTE is a family of equations rather than one universal scalar law.** Electrons, holes, phonon branches, gas species, photons, and plasma particles need separate distributions coupled through collision and field terms. Multiband carriers require indices and possible coherent density-matrix extensions. Relativistic gases use covariant phase space; neutral phonons have no electric-force term; photons have creation and absorption. A useful implementation begins by stating particles, dispersion, dimensionality, statistics, and resolved internal states. **Units and phase-space measures must be audited together.** In a continuum crystal, $\int d^3k/(2\pi)^3$ counts states per real-space volume per band, with explicit degeneracy. A discrete mesh replaces the integral with quadrature weights. Collision rates have inverse-time units, while a source in the BTE has occupation per time. A factor of sample volume, $2\pi$, spin, or cell volume can change conductivity by orders of magnitude while leaving the solver numerically stable. ```svg ``` **Fermi's golden rule often supplies microscopic transition rates.** A perturbation couples initial and final states, and the rate contains the squared matrix element times an energy-conserving delta function and available final-state density. Phonon absorption and emission shift energy by $\pm\hbar\omega$ and carry Bose occupation factors. Disorder scattering is often elastic after configurational averaging. Golden-rule rates assume weak coupling and sufficiently long observation for energy selection; strong coupling, coherent dynamics, and broad spectral functions require more general treatments. **Pauli blocking makes fermionic scattering nonlinear in occupation.** A transition from state $i$ to $j$ contributes $W_{i\to j}f_i(1-f_j)$, and the reverse term must also appear. In a nondegenerate semiconductor, $f\ll1$ lets one approximate $1-f\approx1$; near a Fermi surface that shortcut fails. Blocking changes relaxation, mobility, hot-carrier cooling, and electron–electron phase space. Using Fermi–Dirac equilibrium populations with a classical collision operator can violate detailed balance. **Detailed balance identifies the correct equilibrium nullspace.** At thermal equilibrium, every transition network must leave the Fermi–Dirac, Bose–Einstein, or Maxwell–Boltzmann distribution stationary with the bath temperature and chemical potentials allowed by conserved quantities. Pairwise detailed balance is sufficient but stronger than necessary for some systems. Testing $C[f_0]=0$ numerically is essential. A small but systematic equilibrium collision source creates false current, heating, or chemical production over long simulation times. **Collision invariants determine which moments are conserved.** Number-conserving elastic scattering has a vanishing zeroth collision moment. Momentum-conserving carrier–carrier collisions have a vanishing momentum moment even though they rapidly reshape angular distributions. Energy exchanged with a fixed phonon bath need not be conserved within the electron subsystem, but total electron-plus-phonon energy should balance in a coupled model. A relaxation model that drives all moments to prescribed values can accidentally remove conserved quantities. **The H-theorem formalizes irreversible relaxation under suitable kernels.** For the classical Boltzmann collision operator, entropy-like $H=\int f\ln f$ does not increase when microreversibility and molecular-chaos assumptions hold. Quantum versions include fermionic or bosonic entropy factors. Equality characterizes local equilibrium constrained by collision invariants. MIT plasma-transport material emphasizes conservation, positivity, and the H-theorem as collision-operator requirements. Discrete schemes should mimic these properties where possible rather than merely conserve total particle count. **The relaxation-time approximation is useful because it is transparent and dangerous because it is broad.** Replacing $C[f]$ by $-(f-f_0)/\tau$ makes departures decay exponentially in a homogeneous force-free test and permits analytic linear response. A single $\tau$ generally cannot conserve number, momentum, and energy simultaneously unless $f_0$ is chosen from matching moments. Energy-, band-, position-, and direction-dependent relaxation times improve fidelity but still compress in-scattering structure. MIT transport notes explicitly flag limitations for inelastic electron–phonon exchange. **Transport lifetime differs from single-particle lifetime.** Small-angle scattering may strongly broaden a quantum state yet weakly relax current because it barely changes velocity direction. In isotropic elastic transport, a factor like $1-\cos\theta$ weights momentum relaxation, producing $\tau_{tr}$ distinct from the total scattering lifetime. Quantum mobility inferred from oscillation broadening and transport mobility inferred from conductivity therefore need not agree. Substituting one measured lifetime for another corrupts mean free path and conductivity. **Matthiessen's rule is an approximation to combined collision physics.** Adding inverse relaxation times assumes independent mechanisms acting on the same perturbation with compatible angular and energy structure. It can work for simple elastic channels but fail when scattering mechanisms interfere, redistribute carriers into different energy ranges, or produce different nonequilibrium shapes. A mobility curve fitted by inverse-rate addition may hide compensating errors. Full collision operators add before inversion; their effective transport times need not. **Boundary scattering belongs in kinetic boundary conditions when geometry is resolved.** Incoming distributions at a surface can be specularly reflected, diffusely re-emitted, absorbed, transmitted, thermalized, or mixed according to angle and energy. Fuchs–Sondheimer film corrections and Casimir phonon limits reduce this physics into size-dependent coefficients. A local bulk relaxation time cannot reproduce directional boundary memory when mean free path is comparable to feature size. Boundary kernels must conserve probability and, where appropriate, energy and tangential momentum. **Electron–phonon scattering couples two nonequilibrium populations.** Treating phonons as a fixed equilibrium bath is valid when lattice thermalization is fast and electron power is small. Under strong drive, hot phonons accumulate and reduce carrier cooling, so electron and phonon BTEs exchange equal and opposite energy. Matrix elements depend on deformation potentials, polar coupling, piezoelectricity, screening, band overlap, and phonon branch. Broadening a delta function for numerical integration must converge without creating or destroying net energy. **Carrier–carrier collisions thermalize without directly relaxing total momentum in a clean parabolic band.** Electron–electron scattering can rapidly establish a displaced Fermi distribution and redistribute energy, yet translational invariance protects total crystal momentum modulo Umklapp and band effects. Impurities, phonons, boundaries, or multiple bands then relax current. Treating electron–electron scattering as an ordinary momentum-relaxing time can underestimate conductivity while still being necessary for hydrodynamic local equilibrium. **Impurity scattering requires screening and charge-state consistency.** Ionized dopants generate long-range Coulomb potentials whose small-angle divergence is regularized by dielectric screening. Screening depends on carrier density, degeneracy, temperature, dimensionality, and wave vector. The same incomplete-ionization model that sets impurity charge should feed Poisson and scattering. Central-cell corrections matter for short-range details. Fitting an empirical mobility can absorb screening but loses transferability across density and temperature. ```svg ``` **Linear response expands the distribution around equilibrium.** Write $f=f_0+\delta f$ and retain first order in electric field, temperature gradient, chemical-potential gradient, and $\delta f$. The derivative $-\partial f_0/\partial\varepsilon$ selects an energy window near the chemical potential for degenerate carriers. Solving a linearized collision equation gives $\delta f$, from which fluxes follow. Linear response requires perturbations small enough that coefficients do not depend appreciably on the drive and Joule heating remains a higher-order effect. **The Drude conductivity is a special BTE moment result.** For an isotropic parabolic band, uniform steady field, and constant momentum-relaxation time, the relaxation-time solution yields $\sigma=nq^2\tau/m^*=nq\mu$. This familiar expression hides band nonparabolicity, degeneracy, multiple valleys, anisotropic mass, energy-dependent scattering, and boundary effects. Recovering it is a valuable unit test, but fitting every conductor with one Drude $\tau$ does not validate the underlying kinetic distribution. **Conductivity is generally a tensor built from velocities and scattering.** In a band representation, linearized transport integrates products $v_i v_j$, lifetime or inverse collision operator, and $-\partial f_0/\partial\varepsilon$ over states. Crystal symmetry restricts tensor components; magnetic field adds antisymmetric Hall response; anisotropic scattering rotates principal axes. A scalar mobility erases this structure. Onsager–Casimir reciprocity relates coefficients at reversed magnetic field when microscopic reversibility applies. **Diffusion emerges from spatial gradients of the distribution.** A slowly varying local equilibrium with density or chemical-potential gradients drives an odd-in-velocity perturbation and particle flux. In a classical isotropic limit, random-walk reasoning and BTE moments yield $D\sim v^2\tau/d$. Combining electric and chemical driving produces the Einstein relation under the appropriate statistics. Nonlocal kernels replace a local diffusion coefficient when gradient length approaches mean free path. **Thermoelectric coefficients are coupled moments of the same solution.** Electrical conductivity weights state transport near the chemical potential; the Seebeck coefficient weights particle–hole asymmetry by energy offset; electronic thermal conductivity weights its square after enforcing the electrical constraint. MIT kinetic thermoelectric notes derive these coefficients from BTE structure. Treating them as independent tables can violate Onsager relations and energy balance. Energy-dependent scattering and band structure control deviations from simple Wiedemann–Franz behavior. **The Seebeck coefficient is sensitive to spectral asymmetry rather than conductivity magnitude alone.** Contributions above and below chemical potential transport opposite entropy signs. A flat transport distribution around the Fermi level gives small thermopower even with high conductivity. Band edges, resonances, valley convergence, and energy-filtering barriers can increase asymmetry while reducing conductance. Contact thermopower and bipolar conduction must be included when comparing a device voltage to bulk coefficients. **Thermal conductivity separates electronic and phononic kinetic problems.** Electron BTE moments give electronic heat flow under the condition of zero electrical current for standard measurement. Phonon BTE uses mode heat capacities, group velocities, and scattering lifetimes to give lattice heat flow. Electron–phonon coupling exchanges energy between subsystems. Adding independently calibrated conductivities is valid only if their temperatures are locally locked or coupled consistently; nanoscale hotspots can require two-temperature transport. **The Wiedemann–Franz law is a limiting relation rather than an identity.** Degenerate electrons with elastic scattering varying slowly near the Fermi level give $\kappa_e/(\sigma T)$ near the Sommerfeld Lorenz number. Nondegenerate semiconductors, inelastic scattering, bipolar transport, energy filtering, and strong energy-dependent lifetimes alter it. Subtracting lattice thermal conductivity from total data using a universal Lorenz number can therefore create false trends. **Hall transport reveals angular and energy details hidden by longitudinal conductivity.** Magnetic streaming deflects the nonequilibrium distribution in momentum space. The Hall coefficient equals $1/(qn)$ only in simple single-band limits; Hall factors depend on scattering and statistics, while multiple carriers can change sign and field dependence. Magnetoresistance and Hall mobility constrain collision models more strongly than zero-field conductivity alone. Weak-field linearization fails when cyclotron motion during a relaxation time is not small. **Frequency-dependent response probes relaxation spectra.** With harmonic drive, the time derivative introduces $-i\omega\delta f$. A single relaxation time yields the Drude factor $1/(1-i\omega\tau)$, but realistic collision operators produce multiple modes and memory. Causality connects real and imaginary response through Kramers–Kronig relations. At high frequency, interband quantum transitions and displacement current may lie outside a semiclassical intraband BTE. **Nonlinear high-field BTE predicts distribution heating and velocity saturation.** A strong electric field displaces and distorts $f$ beyond linear response; inelastic phonon emission limits energy and drift velocity. Intervalley transfer can change effective mass and create negative differential mobility, as in the Gunn effect. Assigning a local hot-electron temperature is useful only if carrier–carrier collisions establish a near-thermal shape. Full distributions reveal streaming, tails, and anisotropy that one temperature cannot. **The moment hierarchy explains continuum transport closures.** Integrating BTE with weights $1$, momentum, and energy yields continuity, momentum-balance, and energy-balance equations. Each equation introduces a higher moment such as stress or heat flux, creating an unclosed hierarchy. Drift–diffusion assumes momentum relaxes algebraically and carriers remain near local equilibrium; hydrodynamic models retain momentum or energy; kinetic BTE retains the distribution. Closure quality depends on the ratio of relaxation to macroscopic scales. **Drift–diffusion is a controlled reduction of BTE only under declared assumptions.** Strong momentum relaxation, weak nonlocality, local carrier statistics, and compatible Einstein relations collapse kinetic current to mobility times electrochemical-potential gradient. This explains why the dedicated drift–diffusion page focuses on conservation, contacts, and nonlinear Poisson coupling, while the BTE page focuses on distribution and collisions. Using a field-dependent mobility fitted from BTE can extend the reduction empirically but does not restore directional memory. **Hydrodynamic regimes reverse the usual hierarchy of relaxation times.** If momentum-conserving carrier–carrier or normal phonon scattering is faster than momentum-relaxing impurity, Umklapp, and boundary processes, a drifting local equilibrium forms and collective viscous flow can emerge. Poiseuille profiles, vortices, and second sound are kinetic consequences not described by ordinary diffusion. Boundary slip and sample width become central. A relaxation-time model that damps total momentum cannot capture this regime. **Ballistic transport is boundary controlled rather than collision controlled.** When device length is shorter than relevant mean free paths, incoming distributions are set by reservoirs and propagate with few internal collisions. Landauer transport organizes conduction by transmission modes; a collisionless BTE organizes the same semiclassical limit by characteristics. Contact injection, geometry, and band mismatch dominate. Assigning a local mobility to a ballistic channel makes conductance spuriously proportional to length. ```svg ``` **Knudsen number organizes the departure from local continuum closure.** $Kn=\lambda/L$ compares mean free path with a characteristic geometry or gradient length. Small $Kn$ supports diffusion or hydrodynamic expansions; order-one $Kn$ creates boundary layers and nonlocal response; large $Kn$ approaches free streaming. Real materials have broad mode-dependent mean free paths, so no single $Kn$ describes every carrier or phonon. Suppression functions and cumulative conductivity spectra expose which modes a structure filters. **The Chapman–Enskog method derives constitutive laws from scale separation.** Expand around local equilibrium in a small Knudsen parameter while enforcing that conserved moments reside in the leading distribution. Solvability conditions yield Euler behavior first, then viscosity, diffusion, and heat conduction. Burnett-order corrections can become unstable and are not automatically better. The expansion clarifies why transport coefficients are inverse collision-operator moments and why boundary layers require separate kinetic matching. **The diffusion approximation retains the lowest angular harmonics.** In nearly isotropic transport, write $f$ as an isotropic part plus a small first angular moment that carries flux. Eliminating the rapidly relaxing anisotropic part produces a diffusion equation. The $P_1$ approximation used in radiation, neutron, phonon, and carrier transport embodies this step. It fails near collimated sources, absorbing boundaries, sharp interfaces, or ballistic fronts where higher angular moments remain large. **Spherical-harmonic expansions resolve angular structure systematically.** Expand directional dependence in harmonics and project BTE into coupled equations for coefficients. Low order is efficient near isotropy; high order captures anisotropic scattering and fields but increases memory and can suffer Gibbs oscillations around beams. Crystal momentum space is not generally spherical, so band-adapted meshes or symmetry bases may be preferable. Truncation convergence should be demonstrated on current and energy flow, not coefficient norm alone. **Monte Carlo solves the kinetic problem through sampled trajectories and events.** Ensemble semiconductor Monte Carlo advances particles under band dynamics, samples free-flight times from total rates, selects scattering mechanisms, and accumulates moments. It naturally represents nonlocal and high-field behavior but carries statistical noise and rare-event difficulty. Self-consistent fields require charge deposition and Poisson solves. Null-collision methods simplify variable rates, while variance reduction must preserve unbiased observables. Time step, particle number, mesh, random seeds, and confidence intervals all belong in convergence evidence. **Deterministic discrete-ordinates methods trade noise for phase-space dimensionality.** Discretize directions, energies or wave vectors, positions, and time; then solve coupled advection–collision equations. They provide smooth low-noise distributions and systematic quadrature refinement but face enormous memory and ray effects. Positivity-preserving upwind fluxes, conservative collision integration, and scalable preconditioning are central. Tensor-product grids become impractical for full bands, motivating adaptivity, low-rank methods, sparse grids, or symmetry reduction. **Direct simulation Monte Carlo targets dilute molecular gases rather than electron bands.** DSMC alternates free molecular motion and stochastic collisions within cells, approximating the nonlinear Boltzmann collision integral when cells and time steps resolve mean free path and collision time. It differs from ensemble carrier Monte Carlo in collision pairing, statistics, and force models. Using the name “Monte Carlo BTE” without identifying algorithm and particle physics obscures essential validity conditions. **Lattice Boltzmann is a discrete-velocity moment method, not a direct microscopic semiconductor BTE solver.** Carefully chosen populations and collision rules recover target fluid equations through asymptotic expansion. It excels for mesoscopic fluids and complex boundaries but its populations, velocities, and relaxation parameters are constructed for a continuum limit. Superficial equation resemblance does not make its relaxation time a measured electron scattering lifetime. Stability, isotropy, and equation-of-state constraints follow from the chosen lattice. ```svg ``` **Operator splitting separates streaming, force, and collision updates but introduces commutator error.** Advance real-space advection, reciprocal-space acceleration, and collisions in substeps. Strang splitting is second order for sufficiently smooth operators, while stiff collisions may need implicit or exponential integration. Splitting can violate exact steady balance if field and collision terms cancel but are advanced separately. Positivity and conservation must survive each substep or be restored without changing physical moments. **Real-space advection needs conservative, positivity-aware fluxes.** Upwind finite volume follows characteristic direction and conserves cell-integrated occupation but diffuses sharp fronts. High-order reconstruction reduces diffusion yet can overshoot outside physical occupancy bounds. Discontinuous Galerkin offers local conservation and high order with limiters. Boundary inflow data apply only to velocities entering the domain; prescribing the entire distribution on a boundary overconstrains outflow states. **Reciprocal-space advection must respect Brillouin-zone topology and band geometry.** Electric fields translate crystal momentum, magnetic terms curve trajectories, and periodicity identifies opposite Brillouin-zone faces modulo reciprocal lattice vectors. Interpolation between ab initio band and scattering grids can violate energy or symmetry. Near degeneracies, a single smooth band velocity may not exist. Conservative semi-Lagrangian or finite-volume methods should preserve state count while preventing unphysical $f<0$ or $f>1$ for fermions. **Collision integration is often the stiffest numerical component.** Fast elastic scattering can coexist with slow energy relaxation and device transit, producing widely separated eigenvalues. Explicit updates demand tiny time steps; implicit methods require solving dense or structured state-space couplings. Asymptotic-preserving schemes recover diffusion behavior without resolving every collision time. A numerical method that damps quickly is not necessarily physical if it changes the nullspace or transport coefficients. **The collision matrix should encode conservation in its left nullspace.** After linearization and discretization, vectors representing conserved number, energy, or momentum annihilate the collision operator in the appropriate weighted sense. Equilibrium perturbations tied to conserved chemical potentials or temperature form right-null directions. Projection, constrained solves, or pseudoinverses handle singularity. Accidentally regularizing all zero modes produces finite relaxation of a genuinely conserved quantity and changes DC response. **Positivity and Pauli bounds are physical constraints on distribution updates.** Negative $f$ has no probabilistic meaning, and fermionic occupation above one violates exclusion under the selected normalization. Small violations can destabilize nonlinear blocking factors and entropy. Limiters, implicit collision forms, exponential integrators, and variable transforms help, but clipping changes moments and can hide instability. Report the maximum violation and any conservative correction rather than silently truncating values. **Adaptive energy and momentum meshes should follow active transport windows.** At low temperature, $-\partial f_0/\partial\varepsilon$ narrows around the Fermi level; under high field, a long hot tail and satellite valleys become active; inelastic thresholds create sharp features. Refinement indicators can use flux contribution, collision residual, or interpolation error. Mesh adaptation must transfer occupation conservatively and preserve equilibrium. A grid adequate for density may be inadequate for Seebeck coefficient or rare high-energy ionization. **Full-band transport replaces effective-mass simplifications with numerical dispersion.** Density-functional or empirical pseudopotential calculations provide $\varepsilon_n(\mathbf k)$ and wavefunctions for velocities and matrix elements. Wannier interpolation can make dense Brillouin-zone sampling tractable. Band crossings, gauge choices, valley degeneracy, spin–orbit coupling, and interpolation derivatives demand care. Full bands improve realism only if scattering and doping physics receive comparable fidelity; a precise dispersion paired with arbitrary constant lifetime can still mispredict transport. **Self-consistent electrostatics couples BTE to Poisson through charge.** Integrating occupation over states gives spatial electron and hole densities, which enter $\nabla\cdot(\epsilon\nabla\phi)=-\rho$; the resulting field drives reciprocal-space streaming. Charge deposition, Poisson solution, and distribution update must conserve particles and use consistent reference energies. Strong coupling can produce plasma oscillations or stiffness. Under steady bias, contacts set incoming distributions and potential, while global current and field charge must converge together. **Self-consistent phonons couple BTE to heat and mechanical state.** Phonon occupation determines lattice energy and heat flux; temperature or strain changes dispersion and electron scattering; electron power repopulates phonons. A local Fourier temperature may not exist in a ballistic phonon population, though an energy-equivalent temperature can be defined for reporting. Interfaces need mode-dependent transmission rather than only thermal boundary resistance when spectral nonequilibrium matters. **Quantum corrections become necessary when phase coherence or tunneling controls transport.** Semiclassical BTE tracks occupation of localized wave packets and discards off-diagonal coherence. Density matrices, Wigner transport, Kadanoff–Baym equations, and nonequilibrium Green functions retain progressively more quantum structure. A Wigner function can be negative and should not be subjected to classical positivity logic. Coupling quantum regions to BTE reservoirs requires conserving spectral current and avoiding duplicated scattering. **Plasma kinetic equations share structure but not collision details with semiconductor BTE.** Vlasov streaming couples charged distributions to self-consistent electromagnetic fields; Landau or Fokker–Planck operators describe long-range Coulomb collisions; source, ionization, and wall-sheath physics add species exchange. The dedicated plasma page owns those phenomena. Transplanting semiconductor relaxation times into plasma kinetics, or plasma Maxwellian assumptions into degenerate bands, is invalid despite identical left-hand streaming syntax. ```svg ``` Reservoir contacts impose incoming occupations and absorb outgoing carriers. A thermal contact supplies a Fermi–Dirac or Maxwellian distribution with specified temperature, chemical potential, bands, and modes for velocities entering the domain. Outgoing occupation is determined by the interior unless reflection is modeled. Contact resistance can arise from mode mismatch even without bulk scattering. Imposing a local density on all directions destroys this directional distinction and can create artificial backscattering. Interface transmission must conserve flux rather than raw occupation. Across a band, mass, or phonon mismatch, parallel momentum and energy may be conserved or randomized depending on interface disorder. Transmission and reflection probabilities weight normal group velocity and state density so incident flux equals transmitted plus reflected flux absent absorption. Detailed balance ensures no net interface current at common temperature and chemical potential. Acoustic mismatch, diffuse mismatch, thermionic, and tunneling models encode different assumptions. Open boundaries and periodic driving require compatible gauges. Periodic cells under a field can be formulated with reciprocal-space acceleration, affine electrochemical potential, or source terms, but double-counting the drive is easy. An open device needs inflow data on every characteristic entering its boundary. Artificial absorbing layers should remove particles or energy in a declared physical sink. Boundary placement must be tested because a kinetic disturbance can persist over many mean free paths. Variance and discretization errors demand different convergence studies. Monte Carlo uncertainty falls statistically with sample count and must include autocorrelation; deterministic error falls with phase-space grid and approximation order. Both also carry model and quadrature error in scattering rates. Matching two noisy curves does not prove convergence. Report confidence intervals, independent seeds, grid sequences, time-step sequences, and conservation residuals for the same observable. **Manufactured solutions can verify streaming and collision code independently.** Choose a positive bounded $f(\mathbf r,\mathbf k,t)$, apply the numerical operator analytically or symbolically, and define the source that makes it exact. Test real-space advection, force-space advection, boundary inflow, linear collision, and time integration separately before coupling. Nonlinear two-body collisions need symmetry and equilibrium tests in addition. Measure distribution norms and moment errors because a small integrated current can hide large even-part error. Equilibrium nullspace tests are indispensable collision benchmarks. Evaluate the discrete operator on equilibrium distributions over temperatures and chemical potentials, verify conserved collision moments, and perturb along null and decaying eigenmodes. For paired electron–phonon operators, verify equal and opposite energy transfer. For fermions, test blocking near full occupation. These invariants catch indexing, interpolation, broadening, degeneracy, and detailed-balance defects more directly than a final mobility comparison. Analytic limiting cases create a benchmark ladder. Collisionless advection tests characteristics; homogeneous relaxation tests exponential decay; constant-$\tau$ parabolic bands test Drude conductivity; weak magnetic field tests Hall response; gray phonons test ballistic-to-diffusive slab suppression; a two-state transition tests detailed balance. Progressing from these limits to full bands localizes error and prevents a complex result from becoming its own reference. Global balances must close for every exchanged quantity. Integrate BTE over phase space to compare particle storage, boundary flux, sources, and collision production. Weight by energy and momentum to test their balances, including work by fields and exchange with baths. In steady electrical transport, terminal power should equal internal dissipation plus exported heat under the model. Conservation exactness alone does not guarantee the correct distribution, but imbalance invalidates downstream coefficients. ```svg ``` | Modeling decision | Physical content | Common failure | Decisive check | |---|---|---|---| | distribution normalization | occupation and phase-space state count | missing spin, valley, volume, or $2\pi$ factor | integrate a known equilibrium density | | semiclassical band motion | group velocity and field acceleration | parabolic mass used beyond its energy range | compare full-band velocity and active window | | relaxation-time approximation | exponential return toward chosen equilibrium | conserved moments damped spuriously | collision-nullspace moment test | | golden-rule scattering | weak-coupling transitions | broadened delta violates energy balance | broadening and mesh convergence | | Pauli blocking | finite fermion state availability | classical collision kernel at degeneracy | detailed balance for Fermi–Dirac $f_0$ | | Monte Carlo solution | sampled characteristics and events | noise interpreted as physical structure | independent-seed confidence interval | | discrete ordinates | deterministic phase-space grid | ray effects or negative occupation | angular, energy, and spatial refinement | | reservoir boundary | incoming distribution from a contact | all velocity directions prescribed | equilibrium zero-current contact test | | self-consistent Poisson | field–charge feedback | kinetic and electrostatic gauges mismatch | charge and terminal-current closure | | BTE-to-diffusion reduction | local near-equilibrium closure | used at order-one Knudsen number | compare mean-free-path spectrum with geometry | Validation should target several independent moments and regimes. Electrical conductivity alone can be matched by rescaling scattering. Hall factor probes angular physics, Seebeck coefficient probes energy dependence, thermal conductivity probes energy transport, transient response probes relaxation spectrum, and high-field velocity probes inelastic tails. Compare across temperature, carrier density, crystal direction, magnetic field, and geometry. Reserve some observations from parameter fitting so validation remains predictive. Inverse extraction of scattering rates is often nonunique. Many combinations of band mass, impurity screening, phonon coupling, defect density, boundary specularity, and carrier density reproduce one mobility curve. Spectroscopic linewidths constrain single-particle lifetime but not directly transport lifetime. Joint inference from conductivity, Hall, thermopower, thermal conductivity, and size dependence improves identifiability. Regularization and priors should be reported because a smooth extracted $\tau(\varepsilon)$ may reflect assumptions more than data. Uncertainty is concentrated in collision inputs as much as numerical solution. Matrix elements, impurity densities, phonon dispersions, interface roughness, and band energies carry experimental and first-principles uncertainty. Propagate them to currents and coefficients after numerical error is controlled. Correlated uncertainties matter because the same dielectric function affects screening and polar coupling. A high-precision phase-space solve cannot make poorly constrained rates predictive. Performance claims must include collision construction and observable convergence. Sparse streaming may be cheap while assembling or applying dense scattering dominates. Monte Carlo event tables, full-band interpolation, Poisson coupling, preconditioner setup, communication, and sampling all count. GPU throughput on particles or grid cells is not end-to-end time. Compare methods at equal error and confidence for the requested moment, not equal iteration count or nominal grid size. Reduced-order models need invariant-preserving training and deployment. Low-rank tensor decompositions, neural operators, spectral bases, and surrogate collision maps can compress phase space. Training loss on $f$ may underweight small odd components that carry current. Enforce positivity, state bounds, detailed balance, conservation, symmetries, and correct equilibrium explicitly or through architecture. Validate outside training fields, temperatures, geometries, and scattering mixtures; a surrogate that interpolates current can still predict an unphysical distribution. ```flowchart Declare species, bands or dispersion, statistics, degeneracies, phase-space measure, and observables -> Write semiclassical trajectories, external and self-consistent forces, sources, and boundary inflow -> Build collision kernels with selection rules, blocking or stimulation, and bath exchange -> Verify equilibrium nullspace, detailed balance, positivity, and conserved collision moments -> Compare device, gradient, mean-free-path, and relaxation scales to choose kinetic or reduced model -> Select Monte Carlo, characteristics, discrete ordinates, harmonics, moments, or hybrid discretization -> Converge spatial, angular, energy, band, time, quadrature, broadening, and statistical dimensions -> Close Poisson, phonon, thermal, optical, and contact couplings with global balances -> Recover analytic relaxation, Drude, Hall, diffusion, ballistic, and thermoelectric limits -> Validate multiple independent moments across temperature, density, field, direction, and size -> Archive rates, bands, meshes, seeds, tolerances, hashes, balances, and uncertainty assumptions ``` The most useful diagnostic split is streaming, collisions, boundaries, coupling, numerics, and measurement. A displaced distribution moving the wrong way suggests charge or band-velocity sign. Equilibrium entropy production suggests detailed-balance failure. Correct density with wrong current suggests angular resolution or transport lifetime. Correct bulk conductivity with wrong film size trend suggests boundary scattering. Stable current with failed energy balance suggests inelastic bookkeeping. Parameter retuning should follow, not precede, these invariant checks. | Symptom | Likely source | Targeted investigation | |---|---|---| | $f_0$ is not stationary | collision detailed balance or quadrature | evaluate every in/out pair and weighted nullspace | | particle count drifts | boundary orientation, source, or collision asymmetry | integrated zeroth-moment balance | | conductivity depends on reciprocal-grid rotation | angular quadrature or band interpolation | symmetry-equivalent mesh comparison | | high-field tail changes with energy cutoff | active phase space truncated | expand energy domain and monitor boundary flux | | Monte Carlo mobility varies across seeds | insufficient effective samples | autocorrelation-aware confidence interval | | deterministic solution becomes negative | advection or collision timestep instability | positivity-preserving refinement and limiter audit | | heat and electrical power do not close | energy weights or bath exchange missing | energy-moment balance by mechanism | | ballistic conductance scales with length | diffusive closure or contact model misapplied | collisionless reservoir benchmark | The equilibrium-distribution test should span degenerate and nondegenerate limits. At high temperature and low occupation, Maxwell–Boltzmann behavior provides simple exponential checks. Near a Fermi surface, Pauli factors and a narrow transport window stress energy quadrature. Bosonic phonons test stimulated terms and low-frequency occupation. Evaluating only one room-temperature state can let a collision kernel pass through accidental cancellation. The homogeneous relaxation test isolates temporal integration. With streaming and sources disabled, perturb a known collision eigenmode and measure its decay. A relaxation-time model should give a single exponential; a full operator gives a spectrum and possibly conserved plateaus. Time-step refinement should recover rates without changing null modes. This test separates stiff integrator error from spatial advection and boundary effects. The collisionless slab test isolates kinetic boundaries. Inject known distributions from left and right reservoirs, propagate along characteristics, and compare density and current with analytic velocity integrals. Specular reflection preserves tangential momentum; diffuse reflection resets angular memory; absorbing walls remove flux. This test catches incoming/outgoing orientation and velocity-weighted flux errors before scattering is enabled. The gray phonon slab is a compact ballistic-to-diffusive benchmark. One speed, heat capacity, and mean free path simplify the spectral problem while retaining boundary suppression. In thick samples Fourier conduction emerges away from boundary layers; in thin samples contact temperature jumps and ballistic conductance appear. Matching both limits tests whether a scheme is asymptotic preserving rather than accurate in only one regime. The constant-field parabolic-band test separates linear and nonlinear response. At weak field, the displaced distribution yields Drude conductivity. Increasing field without inelastic energy loss has no physical steady thermal state in an unbounded band; a numerical steady solution then signals artificial cutoff or damping. Adding a bath collision establishes power balance and reveals heating. This prevents a solver boundary in energy space from masquerading as velocity saturation. For semiconductor device work, BTE and Poisson must share band-edge and charge references. Electrostatic potential shifts electron energies with the carrier charge sign; contact chemical potentials set incoming occupation; doping and bound charge set Poisson sources. A global potential gauge shift accompanied by every energy reference should leave observables unchanged. This gauge-invariance test catches silent electron-volt, volt, and sign mismatches. For thermoelectric work, the definition of heat current must subtract electrochemical work consistently. Energy current, heat current, and electrical power are distinct moments. Seebeck measurement imposes zero net electrical current, while thermal conductivity may be reported at zero electric field or zero current. Using coefficients from different constraints breaks the Onsager matrix. The boundary reservoirs also carry Peltier heat that cannot be assigned solely to bulk Joule dissipation. For phonon work, normal and Umklapp processes play different momentum roles. Both conserve energy, but normal scattering preserves crystal momentum within the Brillouin zone while Umklapp transfers a reciprocal lattice vector and relaxes heat current. A single lifetime can reproduce total linewidth while misrepresenting hydrodynamic conductivity. Isotope, boundary, defect, and electron–phonon scattering add further mode dependence. Size and temperature sweeps help distinguish them. For plasma work, long-range Coulomb collisions favor Fokker–Planck structure over binary hard-sphere intuition. Drag and velocity-space diffusion must satisfy a fluctuation–dissipation relation to produce Maxwellian equilibrium. Numerical cutoffs and field-particle terms control conservation. Those details belong to a species- and plasma-specific collision operator, reinforcing why the generic BTE syntax is only the start of a model. For rarefied gas work, the nonlinear collision integral couples pairs of incoming and outgoing velocities under momentum and energy conservation. Molecular chaos factors the two-particle distribution and is an assumption, not an algebraic identity. Internal rotational, vibrational, and reactive states expand the collision kernel. BGK-like relaxations are useful reductions but need corrected Prandtl number or multiple relaxation channels for quantitative heat and momentum transport. Reproducibility requires the complete kinetic ledger: distribution normalization; dimensions; species, bands, and degeneracies; dispersion and velocity interpolation; force and field conventions; every scattering matrix element and broadening; statistics and blocking; contact and surface kernels; phase-space meshes and weights; time integration; random seeds and variance estimates; coupling iteration; conservation tolerances; and observable definitions. A reported relaxation time and grid count cannot reconstruct a BTE calculation. The final interpretation should distinguish what the distribution reveals from what its moments hide. Density is mostly the even component of $f$; current is a small odd component; heat flow weights energy asymmetry; scattering lifetime describes state decay; transport lifetime describes flux decay; and local temperature exists only when the distribution is close to a thermal family. Plotting only density can make a severely nonequilibrium transport state look ordinary. Read the Boltzmann transport equation through a phase-space-streaming-collision-and-moment lens rather than a relaxation-time-and-mobility-formula lens.