Moment-Based Velocity Distribution

• Moment-Based Velocity Distribution is a finite-dimensional kinetic representation that encodes complex velocity fields through a finite set of moments, capturing transport and non-equilibrium features.
• It employs various closure strategies—such as Poisson-EQMOM, regularized entropy, delta quadrature, and Grad methods—to reconstruct distributions while ensuring realizability and convergence.
• These methods are applied in fields like active matter, rarefied gas dynamics, stellar systems, and plasma propulsion to achieve computational efficiency and accurate non-equilibrium modeling.

Moment-based velocity distribution denotes a class of kinetic and statistical representations in which a velocity distribution function is encoded through a finite set of velocity moments and completed by a closure or reconstruction rule. In this setting, the primary unknown is not the full distribution $f$ itself, but quantities such as $\int m(v)\,f\,dv$, Fourier moments on the unit circle, central kinetic moments built from peculiar velocity, or quadrature weights and nodes. Across kinetic equations, active matter, rarefied-gas dynamics, turbulent aerosols, plasma propulsion, and stellar dynamics, the common objective is to reduce an infinite-dimensional velocity description to a finite set of fields while retaining transport, anisotropy, non-equilibrium structure, and, in some formulations, positivity, entropy dissipation, or realizability (Alldredge et al., 2018, Alldredge et al., 2021, Chen et al., 2023, Lin et al., 24 Feb 2025).

1. Moment representation and the closure problem

The basic construction starts from a velocity distribution $f(t,x,v)$ and a prescribed basis of velocity functions $m(v)=(m_0,\dots,m_N)$. The resulting moment vector is

$u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$

which is the canonical form used in entropy-based moment methods (Alldredge et al., 2021). In gas-kinetic formulations, one also distinguishes raw moments,

$$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$

from central moments,

$$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$

so that thermal fluctuations are separated from bulk motion (Lin et al., 24 Feb 2025). For constant-speed particles on the unit circle, the natural variables are complex or Fourier moments,

$$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$

with $m_{-k}=\overline{m_k}$ by symmetry (Chen et al., 2023).

This finite-dimensionalization does not remove the hierarchy. In collisionless transport and in collisional Boltzmann models alike, taking moments produces a chain in which the equation for a moment of order $n$ contains moments of order $\int m(v)\,f\,dv$0. That structure is stated explicitly for stellar systems, where “the equation for the moment of order $\int m(v)\,f\,dv$1 contains moments of order $\int m(v)\,f\,dv$2,” and for collisionless transport, where $\int m(v)\,f\,dv$3 (Schneider et al., 2010, Chalons et al., 2010). Moment-based velocity distribution is therefore inseparable from closure: one must reconstruct unresolved moments, posit a distributional ansatz, or regularize the inverse map from moments back to distributions.

A recurring distinction is between moments that describe conserved bulk quantities and moments that encode shape. Density, momentum, and energy govern hydrodynamic transport, while second- and higher-order moments encode pressure anisotropy, stress, heat flux, kurtosis, or multimodality. This distinction is explicit in Grad-type rarefied-gas theory, where the thirteen moments are $\int m(v)\,f\,dv$4, $\int m(v)\,f\,dv$5, $\int m(v)\,f\,dv$6, the deviatoric stress tensor $\int m(v)\,f\,dv$7, and the heat flux $\int m(v)\,f\,dv$8 (Liu et al., 2022, Liu et al., 2024), and in stellar dynamics, where fourth and fifth moments control the high-velocity tail and anisotropy of the reconstructed velocity distribution function (Schneider et al., 2010).

2. Principal closure families

Several closure families recur across the literature.

Framework	Distributional form	Stated feature
Poisson-EQMOM	$\int m(v)\,f\,dv$9	positivity, smooth reconstruction, convergence as $f(t,x,v)$0 (Chen et al., 2023)
Regularized entropy closure	$f(t,x,v)$1	globally defined for all $f(t,x,v)$2, retains hyperbolicity and entropy dissipation (Alldredge et al., 2018, Alldredge et al., 2021)
Quadrature-based delta closure	$f(t,x,v)$3	exact low-order moment matching, multimodal streaming representation (Chalons et al., 2010)
Grad/Hermite closure	Maxwellian times polynomial corrections in peculiar velocity	explicit stress and heat-flux representation (Pekker et al., 2010, Liu et al., 2022)
Central-moment DBM	discrete velocity distribution built from equilibrium central moments	direct access to thermodynamic nonequilibrium related to thermal fluctuations (Lin et al., 24 Feb 2025)

These closures do not solve the same mathematical problem. Poisson-EQMOM and delta quadrature methods reconstruct a distribution from finitely many prescribed moments. Entropy-based methods choose, among all distributions compatible with a target moment vector, the one minimizing a convex entropy, or a regularized variant when exact realizability is relaxed (Alldredge et al., 2018). Grad-type methods instead posit a polynomial perturbation of a Maxwellian, while central-moment discrete Boltzmann models determine a discrete equilibrium by matching central kinetic moments through matrix inversion (Lin et al., 24 Feb 2025).

A common misconception is that moment closure is synonymous with near-equilibrium Maxwellian fitting. The record is broader. Some closures are singular, as in delta quadrature (Chalons et al., 2010); some are smooth and positive on compact velocity manifolds, as in Poisson kernels on $f(t,x,v)$4 (Chen et al., 2023); some produce exponential-family ansatzes (Alldredge et al., 2021); and some deliberately target non-equilibrium observables such as heat flux, stress, or high-order thermodynamic nonequilibrium tensors (Lin et al., 24 Feb 2025).

3. Constant-speed particles and angular velocity distributions on $f(t,x,v)$5

For kinetic equations with velocity of constant magnitude, the distribution is naturally angular. In the two-dimensional setting with $f(t,x,v)$6, $f(t,x,v)$7, the kinetic equation is

$f(t,x,v)$8

For the Vicsek model of overdamped active particles, the collision operator is a weighted Fokker–Planck operator around the von Mises equilibrium $f(t,x,v)$9, and the equilibria are von Mises distributions (Chen et al., 2023).

The Poisson quadrature method of moments, or Poisson-EQMOM, reconstructs the angular distribution by a homoscedastic mixture of shifted Poisson kernels,

$m(v)=(m_0,\dots,m_N)$0

with shared $m(v)=(m_0,\dots,m_N)$1 and nonnegative weights $m(v)=(m_0,\dots,m_N)$2 (Chen et al., 2023). Because

$m(v)=(m_0,\dots,m_N)$3

the first $m(v)=(m_0,\dots,m_N)$4 Fourier moments are matched exactly: $m(v)=(m_0,\dots,m_N)$5

The stated theoretical advantages are unusually strong for a quadrature closure. Theorem 2.1 gives existence and uniqueness of $m(v)=(m_0,\dots,m_N)$6 and of the associated parameters for realizable truncated moment vectors, so the closure is “well-defined for all physically relevant moments.” Theorem 2.2 gives convergence of $m(v)=(m_0,\dots,m_N)$7 to $m(v)=(m_0,\dots,m_N)$8 in $m(v)=(m_0,\dots,m_N)$9, in $u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$0 for any $u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$1, or uniformly, depending on whether $u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$2 is absolutely continuous, Lipschitz, or Lipschitz with derivative of bounded variation, under the standing assumption $u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$3 (Chen et al., 2023). This directly addresses two classical failure modes of moment closures: nonrealizability and lack of convergence of the reconstructed distribution.

The inversion algorithm is also part of the construction. It finds $u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$4 from the smallest eigenvalue of a Toeplitz/Hermitian moment matrix, reconstructs the node angles from orthogonal polynomials on the unit circle, and recovers the weights from a Vandermonde-type system. A “lifting trick” modifies $u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$5 by $u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$6 to improve accuracy and avoid an expensive search for $u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$7; the paper reports that “$u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$8 up to 100 is feasible” (Chen et al., 2023).

The closure has a specific hyperbolic consequence at lowest order. For $u(t,x)=\int_V m(v)\,f(t,x,v)\,dv,$9, the system is strictly hyperbolic for $$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$0, whereas the limit $$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$1, corresponding to a Dirac delta angular distribution, is not hyperbolic and is non-diagonalizable. This is presented as one reason the smooth Poisson closure differs from delta-type closures (Chen et al., 2023).

4. Position-conditioned velocity distributions in active matter

In interacting active-particle systems driven by Gaussian colored noise, the velocity distribution is not autonomous. Under the Multidimensional Unified Colored Noise Approximation, the stationary conditional velocity distribution at fixed positions is multivariate Gaussian,

$$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$2

with covariance

$$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$3

Thus the local velocity statistics are coupled to the Hessian of the interaction potential (Marconi et al., 2015).

This has explicit two-particle consequences in one dimension. For separation $$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$4 and potential curvature $$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$5, the single-particle conditional variances are

$$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$6

and the cross-correlation is

$$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$7

As $$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$8, these reduce to the free-particle values $$M^{(n)}_{\alpha\beta\dots}=\sum_i f_i \xi_{i\alpha}\xi_{i\beta}\dots,$$9 and $$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$0; at small separation, both approach $$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$1, indicating coherent motion dominated by center-of-mass velocity (Marconi et al., 2015).

The global velocity distribution is therefore not Maxwell–Boltzmann. After averaging over positions, it becomes a mixture of Gaussians, and the paper states that it is “generally non-Maxwellian”; the overall variance depends on density and interaction strength (Marconi et al., 2015). In mean field, for homogeneous one-dimensional systems,

$$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$2

with $$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$3, while in the small-$$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$4 regime the variance is related to the pair distribution function $$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$5 through

$$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$6

A central implication is that moment-based velocity descriptions can encode position–velocity coupling directly, rather than treating the velocity distribution as an equilibrium marginal (Marconi et al., 2015).

5. Gas kinetics, entropy closures, and rarefied-flow solvers

In compressible discrete Boltzmann modeling, the central-moment-based discrete Boltzmann method constructs the equilibrium discrete velocity distribution function $$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$7 by matching central kinetic moments of a Maxwell–Boltzmann equilibrium with internal degrees of freedom. The equilibrium targets include

$$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$8

and the method computes $$\tilde M^{(n)}_{\alpha\beta\dots}=\sum_i f_i(\xi_{i\alpha}-u_\alpha)(\xi_{i\beta}-u_\beta)\dots,$$9 through a central-moment transform $$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$0 (Lin et al., 24 Feb 2025). Because the peculiar velocity $$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$1 is built in, the model “directly” provides nonequilibrium effects related to thermal fluctuation, unlike raw-moment DBMs (Lin et al., 24 Feb 2025).

The same paper gives explicit thermodynamic nonequilibrium diagnostics,

$$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$2

interpreted respectively as viscous stress, heat flux, and higher-order nonequilibrium beyond Navier–Stokes. The method was validated on the Sod shock tube, Lax shock tube, a Mach-15 shock wave, two-dimensional sound waves, and Taylor–Green vortex flow; the sound-wave angle test reported exact $$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$3 versus simulation $$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$4, a relative error of $$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$5 (Lin et al., 24 Feb 2025).

Entropy-based moment methods take a different route. Given a convex entropy density $$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$6, the maximum-entropy closure solves

$$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$7

leading to the ansatz

$$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$8

The regularized variant replaces the exact constraint by a quadratic penalty and solves

$$m_k(t,x)=\int_{-\pi}^{\pi} f(t,x,\theta)\,e^{ik\theta}\,d\theta,$$9

so the closure is globally defined even for nonrealizable $m_{-k}=\overline{m_k}$0 (Alldredge et al., 2018, Alldredge et al., 2021). The trade-off is explicit: the ansatz moments generally differ from the target by $m_{-k}=\overline{m_k}$1, but the mismatch is controlled by $m_{-k}=\overline{m_k}$2. The regularized system retains hyperbolicity, an entropy-dissipation law, and rotational invariance under the conditions stated in the paper (Alldredge et al., 2018).

Convergence of the regularized method back to the original entropy-based moment method has been proved by relative entropy under bounded velocity domain, Lipschitz continuity of the original moment solution, and distance from the boundary of realizability. The paper establishes a relative-entropy rate of order $m_{-k}=\overline{m_k}$3 and an $m_{-k}=\overline{m_k}$4 rate of order $m_{-k}=\overline{m_k}$5, with numerical simulations reported as optimal for these rates (Alldredge et al., 2021).

Rarefied-gas solvers based on Grad’s 13 moments sit between continuum closure and explicit kinetic transport. The Grad distribution used in G13-MGKS is

$m_{-k}=\overline{m_k}$6

and the simplified SG13-MGKS on unstructured meshes adopts the Shakhov collision model to recover the correct Prandtl number (Liu et al., 2024, Liu et al., 2022). The SG13-MGKS abstract states that it gives “reasonably accurate computational results at Knudsen numbers below 0.2,” is “two orders of magnitude faster compared to the conventional discrete velocity method,” and on unstructured grids saves “about 4 times the computation time and 3 times the memory cost” relative to the previous G13-MGKS (Liu et al., 2024). This suggests that moment-based velocity distribution is not only a closure principle but also a practical computational compression strategy.

6. Multistream, tail-dominated, and domain-specific extensions

Moment-based velocity distributions are particularly informative when the underlying velocity law is singular, strongly skewed, or far from equilibrium. In the quadrature-based four-moment model for collisionless transport, the velocity distribution is approximated by a finite sum of Dirac deltas,

$m_{-k}=\overline{m_k}$7

For $m_{-k}=\overline{m_k}$8, the closure exactly represents the first four moments, and smooth solutions are equivalent to two decoupled pressureless gas systems in the primitive variables $m_{-k}=\overline{m_k}$9 (Chalons et al., 2010). The same framework also requires a theory of measure solutions and generalized $n$0-shocks when the physical data require more streams than the closure can represent. This is a reminder that quadrature closures can capture particle trajectory crossing, but only at the cost of admitting singular solutions (Chalons et al., 2010).

In turbulent aerosols, the emphasis shifts from single-particle distributions to pairwise relative velocities. At small separations $n$1, the universal caustic tail is

$n$2

and the corresponding radial relative-velocity density obeys

$n$3

where $n$4 is the phase-space correlation dimension (Gustavsson et al., 2010). The resulting moment asymptotics,

$n$5

show that moments of the relative-velocity distribution can be dominated by caustic events rather than by the smooth core, which is central for collision-rate prediction (Gustavsson et al., 2010).

Astrophysical moment models adopt yet another reconstruction. For dense stellar systems, the velocity distribution function is reconstructed around a Maxwell–Boltzmann baseline by a Legendre-polynomial expansion in spherical velocity space,

$n$6

with coefficients expressed in terms of moments up to fifth order (Schneider et al., 2010). The fourth moments $n$7 and fifth moments $n$8 explicitly control the high-velocity tail, anisotropy, and transport of kurtosis. The paper states that these models permit study of “mechanisms in more detail that increase or decrease the number of high velocity stars” (Schneider et al., 2010).

A plasma-propulsion example appears in collisionless Hall thrusters. There, the axial velocity distribution is obtained analytically from the Vlasov equation under steady state and monoenergetic ion creation. The resulting axial VDF is

$n$9

with moments computed by integrating along Tonks–Langmuir characteristics (Boccelli et al., 2020). The accompanying anisotropic moment system closes the heat flux either by $\int m(v)\,f\,dv$00 or by a polynomial-VDF closure; the paper reports that the polynomial closure, especially with $\int m(v)\,f\,dv$01 and an $\int m(v)\,f\,dv$02-based limiter, substantially improves $\int m(v)\,f\,dv$03 and $\int m(v)\,f\,dv$04 relative to the Euler-like choice (Boccelli et al., 2020).

Across these variants, several limits remain explicit rather than hidden. Poisson-EQMOM assumes constant-magnitude velocity on $\int m(v)\,f\,dv$05 and leaves hyperbolicity beyond $\int m(v)\,f\,dv$06 open (Chen et al., 2023). Regularized entropy closures lose exact moment matching, though with controllable error (Alldredge et al., 2018). Grad-based rarefied-flow solvers are effective at moderate Knudsen number but require higher-order closures as non-equilibrium strengthens (Liu et al., 2022, Liu et al., 2024). This suggests that moment-based velocity distribution is best understood not as a single method, but as a structured family of finite-dimensional kinetic representations whose strengths and failure modes depend on the geometry of velocity space, the choice of moments, and the closure principle.