Linear Boltzmann Transport: Theory and Applications

• Linear Boltzmann Transport is a kinetic framework that describes the linear evolution of particle distributions through free streaming and collision terms in prescribed media.
• It incorporates deterministic, stochastic, and generalized formulations to address varying free-path statistics and memory effects in particle transport.
• Applications span neutron transport, radiative transfer, charged-particle interactions, and jet propagation in quark–gluon plasma, with advanced discretization and reduction methods enhancing computational efficiency.

Linear Boltzmann Transport (LBT) denotes a class of kinetic transport equations and transport frameworks in which a dilute particle or parton distribution evolves by free streaming and linear gain–loss collision terms in a prescribed background medium. In the mathematical, neutron-transport, radiative-transfer, and stochastic-media literature, the term refers to the linear Boltzmann equation and its generalized or non-classical variants; in heavy-ion physics, the same acronym commonly denotes the Linearized Boltzmann Transport model for energetic partons and jets propagating through quark–gluon plasma (Marklof et al., 2015, He et al., 2015). Across these usages, the defining structure is linear evolution of a one-particle phase-space density, with medium properties entering through cross sections, kernels, or externally specified thermal backgrounds rather than through nonlinear evolution of the transported perturbation itself.

1. Canonical transport structure

In deterministic transport theory, a standard steady poly-energetic linear Boltzmann equation on $\Omega_x\times\Omega_\mu\times\Omega_E$ is

$$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$

with inflow boundary condition $u=g_D$ on $\Gamma_-$, where $\alpha$ is the macroscopic absorption cross section, $\beta$ the macroscopic scattering-out cross section, and $S[u]$ the scattering-in operator (Houston et al., 2023). This form makes explicit the standard transport decomposition: streaming along characteristics, attenuation out of the current state, and repopulation from other directions and energies.

A charged-particle approximation used in radiation transport extends the phase space to $G\times S\times I$ and adds explicit energy-loss and angular-diffusion terms: $$a(x,E)\,\frac{\partial \psi}{\partial E} +c(x,E)\,\Delta_S\psi +d(x,\omega,E)\cdot \nabla_S\psi +\omega\cdot \nabla_x\psi +\Sigma(x,\omega,E)\psi -K_r\psi =f.$$ Here $\omega\cdot\nabla_x\psi$ is the streaming term, $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$0 is a slowing-down term, $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$1 encode angular diffusion and drift, $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$2 is attenuation, and $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$3 is a bounded partial integral collision operator (Tervo, 2021). In this formulation, the characteristic inflow boundary value problem is completed by $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$4 and the terminal energy condition $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$5.

These examples illustrate the common operator architecture of LBT. The unknown is a phase-space density or angular flux; the collision operator is linear in that unknown; and the background medium enters through cross sections, scattering kernels, or coefficients that are externally prescribed. This architecture is broad enough to cover neutron transport, radiative transfer, semiconductors, ocean wave scattering, charged-particle transport, and several generalized transport settings (Marklof et al., 2015).

2. Markovian assumptions, kinetic limits, and generalized transport

The classical linear Boltzmann equation relies on an effective Markov property in ordinary phase space. In the standard picture, collisions are effectively uncorrelated, free flights are exponentially distributed, and the process closes on $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$6. In correlated media this fails: the distribution of free path lengths is non-exponential, and the next collision law depends on geometric information inherited from previous flights or collisions (Marklof et al., 2015).

A rigorous example is transport in polycrystals. In that setting, the correct kinetic description is a generalized linear Boltzmann equation on the extended phase space $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$7, where $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$8 is the remaining distance to the next collision and $$\mu\cdot \nabla_{\mathbf x} u + (\alpha+\beta)u = S[u] + f,$$9 is the post-collision velocity. The governing equation is

$u=g_D$0

with a transition kernel that retains memory of previous collision geometry (Marklof et al., 2015). The key transport-theoretic result is that polycrystalline structure still invalidates classical LBT, but the incommensurability of grains yields exponentially decaying free-path distributions, in contrast with the power-law tails found in a single crystal.

A second generalized formulation introduces path length $u=g_D$1 explicitly. In multidimensional stochastic media, the generalized linear Boltzmann equation for the pathlength-dependent angular flux $u=g_D$2 is

$u=g_D$3

with a non-classical total cross section $u=g_D$4 determined by free-path statistics rather than by a constant attenuation rate (Frankel, 2019). This formulation is deterministic but encodes medium stochasticity through pathlength dependence.

The same distinction appears in the velocity-jump derivation of a linear Boltzmann-like equation with memory for arbitrary free-path distributions. There the survival law $u=g_D$5, free-path density $u=g_D$6, and memory kernel $u=g_D$7 replace the classical constant cross section, and the collision terms become nonlocal convolutions in distance or time (Rukolaine, 2015). A direct implication is that non-exponential attenuation is not a perturbation of classical LBT but a structurally different kinetic regime.

Rigorous kinetic-limit derivations further delimit the Markovian regime. For a Lorentz gas in a constant magnetic field, a low-density limit with truncated inverse power-law obstacle potentials yields a linear Boltzmann equation with magnetic transport term,

$u=g_D$8

where $u=g_D$9 is a linear Boltzmann collision operator in velocity space (Marcozzi et al., 2015). The same work shows that hard-disk models can retain non-negligible memory effects in the Boltzmann–Grad limit, so Markovian LBT is not automatic even when the microscopic dynamics are dilute.

3. Charged-particle transport and singular-collision formulations

For charged particles, the exact linear Boltzmann transport operator can be substantially more singular than the bounded partial integral operators familiar from neutral-particle transport. In the Møller-scattering setting, the differential cross section has the structure

$\Gamma_-$0

so the collision operator contains Hadamard finite part integrals of orders one and two (Tervo, 2018). The corresponding exact transport operator is therefore a partial hyper-singular integro-differential operator, not merely a standard bounded collision integral.

The singular energy dependence can be isolated through operators

$\Gamma_-$1

which make the hyper-singular structure explicit (Tervo, 2018). After exact manipulation, the charged-particle transport operator contains first-order energy derivatives, second-order angular derivatives, mixed energy–angle terms, and residual finite-part integrals. This is the analytic origin of CSDA–Fokker–Planck-type approximations.

A reduced charged-particle model replaces the hyper-singular collision operator by partial differential operators plus partial integral operators,

$\Gamma_-$2

posed as a characteristic initial inflow boundary value problem (Tervo, 2021). The associated regularity theory is anisotropic: energy regularity can be arbitrarily high under sufficient smoothness of the data, whereas spatial and angular regularity are limited by characteristic inflow boundaries and by the escape-time geometry. This distinction is important both analytically and numerically, because it shows that isotropic Sobolev intuition is not an adequate guide to charged-particle LBT.

The exact and approximate charged-particle formulations are therefore complementary rather than competing. The hyper-singular equation identifies the true operator structure of charged-particle transport, while the CSDA/Fokker–Planck reduction provides a tractable transport model for applications such as dose calculation in radiation therapy (Tervo, 2018, Tervo, 2021).

4. Linearized Boltzmann Transport for jets and heavy quarks in quark–gluon plasma

In heavy-ion physics, LBT typically denotes the Linearized Boltzmann Transport model for jet propagation in quark–gluon plasma. In its baseline perturbative-QCD formulation, the phase-space distribution $\Gamma_-$3 of a hard parton evolves by elastic scattering on thermal medium partons through a Boltzmann equation that tracks not only the energetic probe but also the thermal recoil partons and the medium depletion generated by removing thermal scattering partners (He et al., 2015). This is the feature that distinguishes the model from simpler jet-energy-loss approaches that evolve only the leading parton or encode the medium by effective drag or $\Gamma_-$4.

The early elastic implementation is restricted to a static, homogeneous QGP and includes the complete set of leading-order pQCD $\Gamma_-$5 channels among quarks, antiquarks, and gluons, regularized by a Debye screening mass. Time evolution is discretized in small steps, the number of scatterings is sampled from a Poisson law, and energy–momentum is conserved exactly in each collision (He et al., 2015). A central bookkeeping structure is the triplet of leading partons, recoil partons, and negative partons. Recoil and negative partons together encode jet-induced medium excitation and the corresponding back reaction; negative partons are propagated and later subtracted from observables.

This linearization has a specific meaning: the thermal background is externally specified, and interactions among jet shower partons and among the jet-induced medium partons themselves are neglected. The induced disturbance is treated as a dilute perturbation on top of a thermal bath characterized locally by $\Gamma_-$6 and fluid velocity $\Gamma_-$7 (He et al., 2015). In this elastic setting, the model computes transverse momentum broadening, elastic energy loss, jet-induced medium excitation, jet shape modification, and fragmentation-function changes, with recoil contributions found to have significant influences on reconstructed-jet observables (Luo et al., 2015).

Later heavy-flavor LBT extends this framework to include both elastic and inelastic processes for light and heavy partons in an expanding hydrodynamic medium. The transport equation is written as

$\Gamma_-$8

with explicit elastic $\Gamma_-$9 scattering and higher-twist medium-induced gluon radiation. Heavy-quark evolution is embedded in viscous hydrodynamics, and hadronization is handled with a hybrid fragmentation-plus-coalescence model (Cao et al., 2016). Within this framework, the model calculations show good descriptions of the $\alpha$0-meson suppression and elliptic flow observed at the LHC and RHIC, while also indicating that a low-momentum enhancement and especially a near-$\alpha$1 enhancement of the transport strength are needed for simultaneous $\alpha$2 and $\alpha$3 phenomenology (Cao et al., 2016).

QLBT is a controlled extension for heavy quarks in which the ideal gas of massless thermal partons is replaced by a quasi-particle medium with temperature-dependent masses constrained by lattice-QCD thermodynamics. The transport equation retains the LBT structure with elastic and inelastic collision integrals, but screening, densities, and thermal occupation factors are modified through the quasi-particle model (Liu et al., 2021). Combined with a hybrid fragmentation–coalescence hadronization approach and Bayesian calibration to $\alpha$4-meson $\alpha$5 and $\alpha$6, QLBT extracts $\alpha$7 and $\alpha$8 in the range $\alpha$9 and compares them with lattice-QCD results and other phenomenological studies.

A more recent improvement modifies the interface between vacuum showering and in-medium transport. Instead of starting transport only after PYTHIA showering reaches the hadronization scale, the model introduces a medium scale $\beta$0 at which in-medium transport is inserted into the vacuum shower and later resumed after the parton exits the QGP (Dang et al., 11 Feb 2026). The same work incorporates color-flow information into the LBT model, enabling string connections correlated with the medium-modified parton shower before hadronization. These two changes alter the predicted ratio of hadron to jet quenching and yield a satisfactory description of the nuclear modification factors of hadrons and jets with different flavors within a unified framework (Dang et al., 11 Feb 2026).

5. Discretization, iterative solvers, and compressed representations

Deterministic LBT computations lead to very large linear systems once space, angle, and energy are discretized. For the stationary poly-energetic linear Boltzmann equation, a tensor-product discontinuous Galerkin finite element discretization in space, angle, and energy gives an operator equation

$\beta$1

where $\beta$2 is the transport operator and $\beta$3 the scattering operator (Houston et al., 2023). In this setting, classical source iteration is a preconditioned Richardson method with preconditioner $\beta$4, and the poly-energetic analogue of the scattering ratio is $\beta$5. The resulting convergence theory and a posteriori solver-error estimates are independent of the discretization parameters, and suitable left/right preconditioning allows standard Euclidean-residual GMRES to produce computable transport-norm error indicators (Houston et al., 2023).

For large transient transport problems, the linear Boltzmann equation also serves as a model reduction target. A full discretization of the time-dependent neutron transport equation yields a linear dynamical system

$\beta$6

which can be reduced either in space only or simultaneously in space and time (Choi et al., 2019). The space-time reduced-order model is built from a separable basis structure and an incremental basis-construction algorithm that is fully parallel and scalable. In the reported neutron-transport examples, the method is applied to problems with $\beta$7 and $\beta$8 space-time unknowns, with wall-clock speedups of $\beta$9 and $S[u]$0, and relative error below $S[u]$1 and below $S[u]$2, respectively (Choi et al., 2019).

High-order geometry-conforming transport discretizations generate a different linear-algebra bottleneck: storage of dense operators. The Tensorized Discontinuous Isogeometric Analysis method treats the steady 2-D linearized Boltzmann transport equation with discrete ordinates in angle, multigroup in energy, and discontinuous isogeometric analysis in space, and assembles the resulting seven-dimensional operators in tensor-train format (Myers et al., 26 Jan 2026). In the reported tests, interior operators can be compressed from petabytes to megabytes, whereas CSR requires gigabytes of storage. The main limitation is that highly coupled boundary operators are difficult for TT, and because the solution vector is left uncompressed, CSR remains $S[u]$3 faster than the mixed TT/CSR format despite the storage savings (Myers et al., 26 Jan 2026).

These developments define a current numerical picture of LBT. High-fidelity discretizations support transport in full space–angle–energy phase space; transport-sweep preconditioning and GMRES provide solver robustness; reduced-order models compress the time dimension as well as the spatial state; and tensor formats can compress high-order operator structure even when they do not yet minimize time-to-solution.

6. Diffusion limits, porous-media transport, and application regimes

A recurring theme in LBT is the passage from mesoscopic transport to macroscopic diffusion. For arbitrary free-path laws, a linear Boltzmann-like equation with memory can be derived from Alt’s model, and when the first and second moments of the free-path distribution are finite, the small-mean-free-path asymptotic solution becomes a diffusion approximation (Rukolaine, 2015). The corresponding diffusion coefficients depend on $S[u]$4 and $S[u]$5, so non-exponential free-path statistics modify macroscopic transport even when a diffusion limit exists.

For stationary generalized linear Boltzmann transport in arbitrary spatial dimension $S[u]$6, rigorous asymptotic diffusion and Grosjean’s moment-preserving construction can be developed for both collision density and scalar flux about an isotropic point source (d'Eon, 2013). The paper shows that Grosjean’s separation of the uncollided distribution from the collided part yields compact analytic approximations that are exact in the first two even spatial moments and are, overall, more accurate for high absorption and for small source–detector separations than either $S[u]$7 diffusion or rigorous asymptotic diffusion (d'Eon, 2013). The same analysis also reveals nontrivial dependence of the transport operator’s discrete spectrum on spatial dimension and on the free-path distribution.

In porous-media column experiments, a one-dimensional linear Boltzmann equation with angular variable $S[u]$8, inherent speed $S[u]$9, bulk drift $G\times S\times I$0, and isotropic scattering coefficient $G\times S\times I$1 is proposed as a mesoscopic model for tracer transport. The measured breakthrough curve is reproduced by the linear Boltzmann equation, while the usual advection–diffusion equation is recovered in the asymptotic limit of large propagation distance and long time (Amagai et al., 2019). In this interpretation, pore branching acts as effective scattering, the transport mean free path is $G\times S\times I$2, and deviations from ADE reflect persistence of mesoscopic angular structure rather than a failure of transport theory itself.

The generalized linear Boltzmann equation also provides a deterministic framework for ensemble transport in bounded, multidimensional stochastic media with finite boundaries. Once the non-classical cross section $G\times S\times I$3 is specified, the resulting pathlength-dependent transport problem can be solved with an implicit discretization of the pathlength variable and standard discrete ordinates sweeps (Frankel, 2019). In the reported Gaussian-process benchmarks, the GLBE matches realization-averaged transport and improves on atomic mix while remaining computationally cheaper than Monte Carlo over realizations (Frankel, 2019).

Taken together, these results place LBT in a hierarchy of kinetic descriptions. At one end lies the classical equation with exponential free paths and Markovian collisions; in the middle are generalized formulations with memory, pathlength dependence, or extended phase space; and at the macroscopic end lie diffusion, CSDA, and Fokker–Planck approximations obtained under problem-specific asymptotic regimes. The appropriate model is therefore determined less by the name “linear Boltzmann transport” than by the medium statistics, collision singularity structure, scale separation, and observables of interest.