Element-Based Limiting Strategy

• Element-Based Limiting Strategy is a family of methods that attach a control variable to each element to enforce physical constraints and avoid spurious oscillations.
• It includes techniques such as a posteriori subcell recomputation, element-wise damping, and convex blending to balance high-order accuracy with robust shock capturing.
• These methods offer guarantees on positivity, entropy stability, and invariant domains, though they may introduce localized dissipation or increased computational cost.

An element-based limiting strategy is a family of numerical stabilization procedures in which the principal control variable is attached to an entire element, cell, or element-local substructure, and is used to suppress nonphysical oscillations or enforce admissibility constraints in high-order discretizations of hyperbolic conservation laws. In discontinuous Galerkin formulations, representative realizations include a posteriori troubled-cell rejection and recomputation on subcells, element-by-element damping of high-order polynomial corrections toward cell averages, and element-wise convex blending between a high-order DG operator and a compatible low-order finite-volume operator. In continuous finite elements, closely related methods are usually algebraic: they act through element-assembled pairwise fluxes or through element-local deviations from admissible cell averages rather than through classical slope limiting. The terminology is not uniform across fields: some papers use “limiting” for constitutive strain saturation embedded in a material law, while other domains use “element-grouping” to denote reduced-resolution control of physical elements rather than shock capturing (Dumbser et al., 2014, Moe et al., 2015, Rueda-Ramírez et al., 2022, Kuzmin et al., 4 Sep 2025, Ghosh et al., 7 Oct 2025, Zhang et al., 2021).

1. Scope, defining objects, and control variables

In the numerical-analysis sense, “element-based” means that the decisive limiter variable is attached to a whole element or to an element-local collection of subcell fluxes. The controlled object may be a troubled-cell flag $\beta_i^{n+1}$, a scalar damping factor $\theta_i$ applied to $q^h-\bar q_i$, a single blending coefficient $\alpha^e\in[0,1]$ combining DG and finite-volume updates on one element, or an element coefficient $\beta_e$ constraining deviations from an admissible intermediate cell average. This contrasts with approaches whose fundamental control variable is a face-based reconstruction, a nodal patch operator, or a global postprocessing step (Dumbser et al., 2014, Moe et al., 2015, Rueda-Ramírez et al., 2022, Kuzmin et al., 4 Sep 2025).

Strategy family	Element-level control variable	Immediate action
A posteriori DG subcell recomputation	$\beta_i^{n+1}$	Reject candidate update and recompute troubled cell locally
Element-by-element DG rescaling	$\theta_i$	Rescale $q^h-\bar q_i$ toward the cell average
DGSEM convex blending	$\alpha^e$ or $\alpha$	Blend high-order DG and compatible low-order FV operators
CG/Bernstein convex limiting	$\theta_i$0	Scale element-local deviation from an admissible cell average

A recurring structural feature is locality. In the a posteriori ADER-DG limiter, detection and recomputation are confined to flagged cells; in the Barth–Jespersen/Zhang–Shu-type DG limiter, one scalar $\theta_i$1 rescales the entire polynomial on the current element; in element-wise DGSEM blending, one constant value of $\theta_i$2 determines the amount of finite-volume content added to all nodal updates in that element; and in matrix-free CG convex limiting, one $\theta_i$3 controls all element-local auxiliary states $\theta_i$4 generated from the same admissible average $\theta_i$5 (Dumbser et al., 2014, Moe et al., 2015, Rueda-Ramírez et al., 2022, Kuzmin et al., 4 Sep 2025).

2. Elementwise strategies in discontinuous Galerkin methods

A canonical element-based DG limiter is the a posteriori subcell technique for ADER-DG. The algorithm computes an unlimited candidate solution $\theta_i$6, tests it cell by cell by physical admissibility detection and numerical admissibility detection, and accepts it only if both tests pass. Physical admissibility is problem-dependent; for the compressible Euler equations the cited constraints are positivity of density and pressure. Numerical admissibility is enforced through a relaxed discrete maximum principle on a neighborhood $\theta_i$7, with tolerance $\theta_i$8 and $\theta_i$9. Troubled cells are not repaired by coefficient clipping. Instead, the element is recomputed from the previous-time DG state projected onto a Cartesian subgrid of $q^h-\bar q_i$0 finite-volume subcells, evolved by a third-order ADER-WENO finite-volume scheme, and reconstructed back into a degree-$q^h-\bar q_i$1 DG polynomial. The design choice $q^h-\bar q_i$2 matches the admissible CFL time step of the subgrid FV method to that of the DG scheme on the main grid, and the projection/reconstruction pair satisfies $q^h-\bar q_i$3 when $q^h-\bar q_i$4 (Dumbser et al., 2014).

A different element-based DG strategy limits the polynomial directly. The high-order DG state is written as $q^h-\bar q_i$5, and the limited state is $q^h-\bar q_i$6. The coefficient $q^h-\bar q_i$7 is computed from local admissible upper and lower bounds inferred from neighboring elements and from sampled extrema of the current element polynomial. The sampling set $q^h-\bar q_i$8 contains interior Gaussian quadrature points, corner points, and Gaussian quadrature points on element boundaries. For a scalar variable, neighbor-based bounds are

$q^h-\bar q_i$9

and the limiter uses the cutoff $\alpha^e\in[0,1]$0. The tolerance $\alpha^e\in[0,1]$1 must vanish more slowly than $\alpha^e\in[0,1]$2 to avoid clipping smooth extrema; the practical choice highlighted in the paper is $\alpha^e\in[0,1]$3. For systems, bounds are computed in primitive variables while the rescaling is applied to conserved variables, preserving conservation exactly through the invariant cell average (Moe et al., 2015).

These two formulations express two distinct notions of element-based control. The first is element-local rejection and recomputation with another solver on a finer internal mesh; the second is elementwise damping of the existing DG polynomial. Both are local, conservative, and designed to preserve high-order accuracy where limiting is inactive, but they attach the limiter to fundamentally different objects: a completed update in one case and the high-order correction in the other (Dumbser et al., 2014, Moe et al., 2015).

3. Convex blending, subcell control, and entropy in DGSEM

In nodal DG spectral element methods, element-based limiting is often expressed as convex blending between a high-order DGSEM operator and a compatible low-order subcell finite-volume operator. A general framework identifies four ingredients: a nodal high-order DG method on Legendre–Gauss–Lobatto nodes, a compatible robust subcell FV scheme, a convex combination strategy that may be element-wise or subcell-wise, and a mechanism for computing blending factors either from troubled-cell indicators or from flux-corrected-transport ideas. For element-wise blending, the defining formula is

$\alpha^e\in[0,1]$4

with one constant $\alpha^e\in[0,1]$5 for the whole element. This formulation is locally conservative because both operators share the same nodal unknowns and a common flux-difference structure. It is also the setting in which the paper cites a proof that any element-wise convex combination of an entropy-stable FV scheme and an entropy-stable DGSEM scheme remains provably entropy-stable, provided the surface terms match. The price is increased dissipation relative to subcell-wise blending, since one problematic node forces the same $\alpha^e\in[0,1]$6 on all nodal updates in the element (Rueda-Ramírez et al., 2022).

Monolithic convex limiting for LGL-DGSEM exploits the fact that the collocated SBP structure yields a natural decomposition into subcell interface fluxes between neighboring LGL nodes. The method introduces a low-order LLF flux, a high-order DGSEM flux in flux-differencing form, and an antidiffusive flux $\alpha^e\in[0,1]$7 between subcell neighbors. Limited bar states

$\alpha^e\in[0,1]$8

are constrained to remain in convex invariant sets, yielding positivity-preserving and local-bound-preserving updates under an SSP/CFL restriction. The construction supports sequential limiting of density, velocity components, and specific total energy, a synchronized pressure-positivity factor, and an optional semidiscrete entropy fix. Its defining feature is monolithicness: the nonlinear spatial fluxes are independent of the time-step size, unlike predictor-corrector FCT variants (Rueda-Ramírez et al., 2023).

Entropy-oriented subcell limiting sharpens this DGSEM line of work by posing the limiter coefficients as solutions of an optimization problem on each element. The target is not merely positivity or a subcell entropy principle but a provable semi-discrete cell entropy inequality. In one dimension, the limiter coefficients $\alpha^e\in[0,1]$9 solve a continuous knapsack-type linear program that maximizes $\beta_e$0 subject to an element entropy budget and convex constraints $\beta_e$1. In multiple dimensions, analogous linear programs are solved per element and per coordinate direction. The low-order method supplies feasibility, the optimization preserves local conservation, and a deterministic greedy algorithm yields the exact LP solution efficiently (Lin et al., 2023).

A later refinement replaces the linear knapsack objective by the quadratic program

$\beta_e$2

The admissible set is unchanged, but the limiter map becomes continuous in the inputs rather than piecewise discontinuous. The paper proves that the resulting quadratic knapsack problem reduces to scalar root finding and reports improved temporal regularity, fewer adaptive timesteps in shock-type problems, and approximately $\beta_e$3 time convergence where the earlier linear-knapsack limiter exhibited approximately $\beta_e$4 behavior (Christner et al., 19 Jul 2025).

The same element-based convex-limiting philosophy has been extended to nonconforming AMR interfaces in LGL-DGSEM. There the new ingredient is an invariant-domain-preserving low-order mortar flux on hanging-node interfaces, built from sparse overlap weights $\beta_e$5 and limited by one mortar coefficient $\beta_e$6. The method preserves conservation, reduces to the conforming formulation on matching interfaces, fits graph-viscosity low-order schemes, and supplies the missing ingredient needed to combine DGSEM, invariant-domain-preserving limiting, and adaptive mesh refinement within one framework (Bolm et al., 7 Jul 2026).

4. Continuous finite elements and algebraic convex limiting

For continuous Galerkin methods, element-based limiting is usually algebraic rather than polynomial. One influential formulation rewrites the semidiscrete CG scheme as a sum of element-assembled pairwise fluxes $\beta_e$7 and then applies monolithic convex limiting to the antidiffusive part. The low-order baseline is an LLF-type AFC scheme written in bar-state form, while entropy stability is enforced through the pairwise inequality

$\beta_e$8

The limiter clips pairwise antidiffusive fluxes so that corrected bar states remain within local bounds $\beta_e$9, and then applies an algebraic entropy fix to the same fluxes. The resulting methodology is parameter-free, conservative, entropy-stable, and local-maximum-principle preserving, but its locality is best described as element-assembled and pairwise rather than as classical elementwise slope limiting (Kuzmin et al., 2020).

High-order Bernstein finite elements support a more explicitly element/subcell-oriented variant. The key construction is a compact-stencil invariant-domain-preserving low-order scheme on the Bernstein control net together with an exact decomposition of each high-order element correction into nearest-neighbor subcell fluxes. The element residual $\beta_i^{n+1}$0 is split into a graph-viscosity part and a zero-sum remainder $\beta_i^{n+1}$1; the latter is represented by antisymmetric pairwise fluxes $\beta_i^{n+1}$2 through the local sparse solve

$\beta_i^{n+1}$3

The resulting target fluxes $\beta_i^{n+1}$4 vanish for non-nearest neighbors, preserving the compact $\beta_i^{n+1}$5-type subcell stencil inside each macroelement. Limiting then acts directly on these nearest-neighbor fluxes. The paper proves local maximum-principle and invariant-domain properties under CFL-like conditions and emphasizes matrix-free implementation with precomputed reference-element data (Kuzmin et al., 2019).

A matrix-free CG framework with nonlinear stabilization pushes the element-local viewpoint further by introducing admissible intermediate cell averages $\beta_i^{n+1}$6 and element-local auxiliary nodal states

$\beta_i^{n+1}$7

Here $\beta_i^{n+1}$8 is the element limiter coefficient, $\beta_i^{n+1}$9 is the old element average, and $\theta_i$0 are antidiffusive element contributions extracted from the high-order stabilized CG scheme. A separate facet-flux limiter $\theta_i$1 may be used to guarantee admissibility of $\theta_i$2 under a weaker macrocell CFL restriction, after which $\theta_i$3 scales only the element-local deviation from that admissible average. The analysis relies on the positivity and partition-of-unity properties of Bernstein basis functions and yields an invariant-domain-preserving SSP-stage update under a timestep restriction (Kuzmin et al., 4 Sep 2025).

Not every continuous-FE limiter described in elementwise language is truly element-based in the same sense. A well-balanced monolithic convex limiter for steady convection–diffusion–reaction equations is fundamentally an assembled algebraic edge-based method. Its novelty is source-aware balancing: source terms are moved into the same pairwise bar-state and flux machinery as convection through balancing fluxes $\theta_i$4, a source contribution $\theta_i$5, and fictitious-node values computed from local element extensions. This preserves a simple constant-coefficient linear equilibrium exactly, but the limiting still acts on internodal pairs rather than on a single element control variable (Knobloch et al., 2024).

5. Terminological non-equivalences and adjacent meanings

A major terminological pitfall is the difference between numerical element limiters and constitutive strain-limiting models. In the thermo-mechanical crack-tip model for transversely isotropic materials, the “limiting” mechanism is entirely embedded in the material law: $\theta_i$6 As $\theta_i$7, the denominator tends to zero, so stress can become very large while strain remains bounded. The finite element method is a standard conforming continuous Galerkin discretization with quadrature-point evaluation of the constitutive nonlinearity and Picard lagging of the nonlinear factor. The paper is explicit that this is not an element-wise limiter in the numerical-analysis sense: it contains no elementwise strain clipping, slope limiting, artificial viscosity, flux limiting, DG limiting, local postprocessing, or adaptive element-level regularization rules (Ghosh et al., 7 Oct 2025).

A related but distinct use appears in finite element approximations of implicit strain-limiting elasticity. There, piecewise constant stresses and continuous piecewise linear displacements allow the monotone nonlinear constitutive update to be solved independently on each element. The paper describes this as an appealing feature because the monotone part can be computed by solving an algebraic system with $\theta_i$8 unknowns independently on each element. This is element-local limiting in a constitutive saturation sense, not shock capturing or flux correction (Bonito et al., 2018).

Outside numerical PDE stabilization, “element-based limiting” can denote coarse-grained control of physical elements. In RIS channel extrapolation with element-grouping, $\theta_i$9 physical RIS elements are partitioned into $q^h-\bar q_i$0 groups of size $q^h-\bar q_i$1, each group shares one reflection coefficient, and pilot overhead drops from $q^h-\bar q_i$2 to $q^h-\bar q_i$3. The price is a grouping error $q^h-\bar q_i$4, interpreted as channel interference, which is then mitigated by two deep-learning networks for interference elimination and full-channel extrapolation. This use of “element-limiting” is therefore a reduced-resolution acquisition strategy rather than a numerical limiter attached to a finite element discretization (Zhang et al., 2021).

6. Guarantees, trade-offs, and recurrent issues

Across the major PDE families, element-based limiting is valued because it can be highly selective while preserving high-order behavior where the limiter is inactive. The a posteriori ADER-DG subcell limiter achieves essentially the theoretical order $q^h-\bar q_i$5 for polynomial degrees $q^h-\bar q_i$6 to $q^h-\bar q_i$7 in a smooth-vortex study; the element-by-element DG rescaling limiter regains asymptotic high-order behavior when $q^h-\bar q_i$8, whereas $q^h-\bar q_i$9 clips smooth extrema; the entropy-constrained DGSEM optimizer preserves high-order accuracy numerically, with rates between $\alpha^e$0 and $\alpha^e$1 in the isentropic vortex test; the quadratic-knapsack entropy limiter recovers the expected spatial order $\alpha^e$2 on a 1D density wave; and LGL-DGSEM monolithic convex limiting preserves the expected high-order convergence when only global positivity bounds are enforced, while tight local bounds degrade accuracy in smooth regions (Dumbser et al., 2014, Moe et al., 2015, Lin et al., 2023, Christner et al., 19 Jul 2025, Rueda-Ramírez et al., 2023).

The strongest formal guarantees concern invariant domains, positivity, and entropy. Element-wise convex blending in DGSEM is especially attractive when provable entropy dissipation is a priority, because the cited proof covers element-wise convex combinations of entropy-stable DG and FV operators with matching surface terms; the same framework also states that element-wise blending is usually more dissipative than subcell-wise blending. In continuous FE, algebraic convex limiting combines local maximum principles and entropy stability through pairwise inequality constraints, while matrix-free CG convex limiting proves invariant-domain preservation by representing each SSP stage as a convex combination of admissible element-local states. For adaptive nonconforming DGSEM, the low-order mortar flux and one-coefficient-per-mortar construction provide the missing invariant-domain-preserving interface mechanism on hanging-node meshes (Rueda-Ramírez et al., 2022, Kuzmin et al., 2020, Kuzmin et al., 4 Sep 2025, Bolm et al., 7 Jul 2026).

A recurrent misconception is that all “limiting” language refers to the same numerical object. Some methods limit completed updates a posteriori, some rescale high-order corrections a priori or stagewise, some solve a small optimization problem to allocate an element entropy budget, and some embed the only “limit” inside a constitutive law. Another recurrent trade-off is localization versus dissipation. Element-wise blending is simpler and cheaper in storage than subcell-wise blending but damps smooth subregions whenever one node in the element requires limiting. Detector-free elementwise rescaling avoids a separate troubled-cell indicator but only enforces bounds at sampled points, not exact continuous extrema. A plausible implication is that the most appropriate strategy depends less on the word “element-based” itself than on the exact level at which admissibility is enforced: completed cell updates, polynomial corrections, subcell fluxes, admissible cell averages, or constitutive response (Moe et al., 2015, Lin et al., 2023, Ghosh et al., 7 Oct 2025).