Moment-Conserving Limiter Overview

• Moment-conserving limiters are numerical postprocessing techniques that enforce admissibility constraints while preserving a prescribed integral moment, such as cell average or total mass.
• They employ methods like Hermite WENO reconstructions, convex optimization, and redistribution limiting to selectively damp high-order oscillations without altering the conserved moment.
• These techniques are crucial in applications ranging from hyperbolic conservation laws to population balance equations, ensuring stability and physical accuracy across diverse numerical frameworks.

A moment-conserving limiter is a limiting or postprocessing procedure that enforces admissibility constraints while preserving prescribed moments of a discrete solution. In many finite-volume and discontinuous Galerkin formulations, the preserved quantity is the zeroth moment, i.e. the cell average or total mass. In other settings the physically relevant invariant is a higher moment: in aggregation and breakage for the population balance equation, local mass corresponds to the first moment, so preserving the cell average is not the correct constraint (Fan et al., 2024, Liu et al., 2023, Xu et al., 22 Jul 2025). Recent work uses the term across several numerical frameworks, including Hermite WENO reconstructions, convex optimization on cell averages, realizability-preserving scaling in moment space, monolithic convex DG limiting, and locally conservative redistribution (Alldredge et al., 2015, Hajduk, 2020, Guermond, 22 Jun 2026).

1. Meaning of “moment-conserving”

The defining feature is the preservation of a selected integral quantity during limiting. For uniform-mesh DG cell averages, the total mass is the unweighted sum of cell averages, and the conservation constraint is

$\sum_{i=1}^N x_i=\sum_{i=1}^N \bar u_i.$

For finite-volume Hermite WENO schemes, the preserved quantity is the zeroth moment $m^{(0)}$, while the first-order moments are modified by a limiter that remains consistent in smooth regions. For the population balance equation, the local particle mass on a cell $I_K$ is

$m_1=\int_{I_K} v\,n(v)\,dv,$

so a limiter that preserves only $m_0$ is not directly applicable (Liu et al., 2023, Fan et al., 2024, Xu et al., 22 Jul 2025).

This suggests that the term is problem-dependent rather than tied to a single canonical formula. The conserved moment is chosen by the underlying conservation law or physical invariant.

Context	Preserved moment	Limiting mechanism
HWENO-U for hyperbolic conservation laws	Zeroth moment $m^{(0)}$	Unified HWENO reconstruction and time-stage moment replacement
Bound-preserving CHNS limiter	Global zeroth moment / mass	Constrained minimization on cell averages
DG for population balance equation	Prescribed moment $M_s$, especially first moment	Positivity-preserving scaling $n_h^{\mathrm{lim}}(v)=\theta(n_h(v)+m)$
Realizability-preserving DG for $M_N$ closures	Cell-average moment	Ray-based scaling of high-order DG content
Monolithic convex DG limiting	Zeroth moment / cell average	Limited antidiffusive corrections with zero elementwise sum
Conservative redistribution limiting	Generalized mass in native form	Cellwise redistribution and weighted averaging

2. Canonical constructions

A standard construction rescales high-order content around a mean state. In the DG scheme for entropy-based moment closures, the limiter follows the ray

$$m(\theta)=\bar m+\theta\big(m_h(x_q)-\bar m\big),\qquad 0\le \theta\le 1,$$

so that only high-order DOFs are damped and the cell average $m^{(0)}$0 is unchanged. In monolithic convex limiting for DG, the scalar analogue is

$m^{(0)}$1

and the flux-based formulation achieves the same conservation property by constructing antidiffusive corrections whose sum vanishes per element and that are antisymmetric across faces (Alldredge et al., 2015, Hajduk, 2020).

A different class formulates limiting as a constrained correction of low-order moments or cell averages. For the CHNS system, the limited cell averages are obtained from

$m^{(0)}$2

which preserves the zeroth moment exactly up to round-off at convergence. In conservative redistribution limiting, one instead builds cellwise provisional states $m^{(0)}$3 satisfying

$m^{(0)}$4

and then assembles the final state by weighted averaging over incident cells (Liu et al., 2023, Guermond, 22 Jun 2026).

These constructions share a common algebraic pattern: the limiter is allowed to change oscillatory or inadmissible high-order content, but not the selected linear invariant.

3. Unified Hermite WENO limiting

The paper “A moment-based Hermite WENO scheme with unified stencils for hyperbolic conservation laws” introduces a fifth-order HWENO-U scheme for scalar and system hyperbolic conservation laws in one and multiple dimensions. In one dimension it evolves the pair

$m^{(0)}$5

with a quartic Hermite polynomial $m^{(0)}$6 reconstructed on the compact stencil $m^{(0)}$7 and linear polynomials $m^{(0)}$8 reconstructed on $m^{(0)}$9 and $I_K$0. The nonlinear HWENO polynomial is

$I_K$1

and a direct calculation shows

$I_K$2

The limited first moment is then defined as

$I_K$3

The same pattern extends to two dimensions with $I_K$4 and a unified reconstruction $I_K$5 on compact $I_K$6 and directional $I_K$7 stencils (Fan et al., 2024).

The distinctive feature is that limiting is applied only to the first-order moments during time stepping. With SSP RK3, the zeroth-moment update and the flux-difference operator are untouched, while the stage values of $I_K$8 are replaced by the HWENO-limited $I_K$9. Therefore conservation of the zeroth moment is preserved exactly, because the update of $m_1=\int_{I_K} v\,n(v)\,dv,$0 remains the standard conservative finite-volume form. The paper explicitly describes this as “moment-conserving” in the sense of preserving $m_1=\int_{I_K} v\,n(v)\,dv,$1 while stabilizing $m_1=\int_{I_K} v\,n(v)\,dv,$2 (Fan et al., 2024).

The scheme also introduces scale-invariant nonlinear weights

$m_1=\int_{I_K} v\,n(v)\,dv,$3

with $m_1=\int_{I_K} v\,n(v)\,dv,$4 constructed only from integral average values of the solution. Under the scaling $m_1=\int_{I_K} v\,n(v)\,dv,$5, the weights are identical for $m_1=\int_{I_K} v\,n(v)\,dv,$6 and $m_1=\int_{I_K} v\,n(v)\,dv,$7, unlike the original scale-dependent forms. The paper states that this normalization has no adverse effect on accuracy and markedly improves robustness for sharp scale variations. A von Neumann analysis for linear advection shows that omitting the time-level modification of the first moment gives unstable fully discrete schemes, while the modified RK3 procedure yields a stable scheme; in the linear case with unified stencils the necessary stability condition is $m_1=\int_{I_K} v\,n(v)\,dv,$8. Extensive 1D and 2D tests show fifth-order convergence of the solution and also fifth-order convergence of the first moments $m_1=\int_{I_K} v\,n(v)\,dv,$9, despite their time-stage limiting (Fan et al., 2024).

4. Convex optimization and redistribution formulations

In the CHNS setting, bound preservation and moment conservation are posed as a nonsmooth convex composite problem. The feasible set is the intersection of the affine conservation manifold

$m_0$0

and the box

$m_0$1

The generalized Douglas–Rachford splitting method is then applied to the split

$m_0$2

with explicit resolvents for the mass hyperplane and the box projection. The paper states that for each time step it takes at most $m_0$3 iterations for the Douglas–Rachford splitting to enforce bounds and conservation up to the round-off error, and that the computational cost is at most $m_0$4 with $m_0$5 the total number of cells. After the cell-average correction, a Zhang–Shu scaling limiter is applied at quadrature points without changing the cell average, so mass and high-order accuracy are maintained (Liu et al., 2023).

A broader redistribution perspective appears in “Locally conservative redistribution limiting and applications to the approximation of conservation equations, Part II”. There the generalized mass is

$m_0$6

and the native algorithm conserves the zeroth moment componentwise in systems. On each cell $m_0$7, the provisional correction satisfies

$m_0$8

and the final limited state is

$m_0$9

The paper emphasizes that the method is non-intrusive, discretization and PDE agnostic, conservative, and applicable to elliptic, parabolic, and hyperbolic problems. It also makes an important distinction: for DG-like discretizations where $m^{(0)}$0, the cell-averaged mass is preserved exactly in every cell, whereas after Jacobi averaging in continuous or spectral settings only global mass is guaranteed (Guermond, 22 Jun 2026).

5. Preserving higher moments and realizability

The population balance equation provides a clear case in which the physically relevant conserved quantity is not the zeroth moment. In aggregation and breakage, local mass corresponds to the first moment, and the paper states that the classical Zhang–Shu limiter, which preserves the zeroth moment, is not directly applicable. The proposed DG limiter assumes a nonnegative target moment

$m^{(0)}$1

defines

$m^{(0)}$2

and sets

$m^{(0)}$3

The construction guarantees positivity, exact preservation of the $m^{(0)}$4-th moment,

$m^{(0)}$5

and no increase of the maximum amplitude. For aggregation and breakage, the relevant choice is $m^{(0)}$6, so the limiter preserves local mass rather than local number. Numerical results reported in the paper show third-order convergence for $m^{(0)}$7 DG and mass conserved to machine precision (Xu et al., 22 Jul 2025).

A different admissibility notion appears in entropy-based moment closures for kinetic equations. The realizability-preserving DG scheme for the $m^{(0)}$8 model uses a numerically realizable convex polytope $m^{(0)}$9 and computes the largest $M_s$0 such that

$M_s$1

remains in $M_s$2 at all required points. The limiter therefore rescales only high-order DG coefficients and preserves the cell-average moment exactly. The paper presents this as a generalization of positivity-preserving linear scaling limiters to high-order moment systems with complex admissibility sets (Alldredge et al., 2015).

For finite-volume transport of moment sets, a related but distinct conservative mechanism appears in “A realizable second-order advection method with variable flux limiters for moment transport equations”. The method reconstructs a realizable moment set at a cell face by allowing flexible selection of flux limiter values inside the second-order TVD region. Conservation is preserved because fluxes are computed per face and applied with opposite signs in adjacent cells, so each component $M_s$3 is advanced by a conservative flux. The paper gives necessary conditions to satisfy realizability and the second-order TVD property for the third-order moment set, and conditionally extends the strategy to the fourth- and fifth-order moments (Choi et al., 2022).

6. Common properties, misconceptions, and open directions

The recent literature shows that “moment-conserving” does not mean “all moments are preserved”. In HWENO-U, the limiter is moment-conserving because it preserves the zeroth moment exactly while modifying the first-order moments only through a convex HWENO projection. In the PBE setting, preserving the zeroth moment can be physically incorrect because the conserved quantity is the first moment. In locally conservative redistribution limiting, the native algorithm preserves only the zeroth moment, and the paper explicitly states that preserving higher moments would require extra linear constraints that it deliberately avoids (Fan et al., 2024, Xu et al., 22 Jul 2025, Guermond, 22 Jun 2026).

A second misconception is that moment conservation by itself guarantees positivity, realizability, or non-oscillatory behavior. The cited schemes obtain these properties through different supplementary mechanisms: scale-invariant WENO weights and time-stage moment replacement in HWENO-U; box projection and hyperplane projection in convex optimization; positivity nodes and moment-preserving scaling in the PBE limiter; ray–polytope intersection in realizability-preserving DG; and invariant-domain bar states or redistribution budgets in convex and redistribution limiters (Liu et al., 2023, Alldredge et al., 2015, Hajduk, 2020).

The implementation trade-offs are correspondingly different. HWENO-U emphasizes a single reconstruction with unified stencils, shared candidate stencils, reconstructed polynomials, and linear and nonlinear weights. The CHNS limiter emphasizes an $M_s$4 first-order splitting algorithm with at most $M_s$5 iterations per step. The realizability-preserving $M_s$6 scheme moves the main complexity into the H-representation of the numerically realizable polytope, whose facet count grows rapidly with moment order and angular resolution. The PBE scheme places its dominant cost in symmetric double-integral evaluation on common refinements, while the redistribution method emphasizes $M_s$7 cell-local loops (Fan et al., 2024, Liu et al., 2023, Alldredge et al., 2015, Xu et al., 22 Jul 2025, Guermond, 22 Jun 2026).

A plausible implication is that the modern notion of a moment-conserving limiter is best understood as a design principle rather than a single algorithmic family: enforce admissibility, alter only the part of the discrete representation that is responsible for nonphysical behavior, and preserve exactly the moment that encodes the governing conservation law.