Active Flux and Polygons: why a stabilisation and how to implement boundary conditions

1 Introduction

• VEM projection onto Pk loses unresolved modes, motivating filtering to damp them without compromising accuracy.The VEM space contains Pk as a proper subspace, unlike the simplex schemes discussed here.Source: §1 · p. 1
• The prior boundary-condition method left treatment of point-value degrees of freedom problematic.Its derivation was highly heuristic and theoretically unsatisfactory.Source: §1 · pp. 1–2
• The note reinterprets PAMPA in a dG framework with projection to justify and improve stabilization and modify boundary treatment.The approach follows [5].Source: §1 · p. 2

2 Meshes and VEM approximation setting

• The mesh polygons are star-shaped, and shape regularity requires the diameter-to-inradius ratio to stay uniformly bounded.The setting assumes non-overlapping polygons in two dimensions for simplicity.Source: §2 · p. 2
• For k ≥ 2, the VEM degrees of freedom include vertex values, internal Gauss–Lobatto edge values, and interior moments.The moment indices satisfy 0 ≤ |α| ≤ k−2.Source: §2 · pp. 2–3
• The VEM space contains Pk, but its L2 projection cannot be computed solely from the original degrees of freedom.The elliptic projection, in contrast, depends only on those degrees of freedom.Source: §2 · p. 2
• An auxiliary space Wk is constructed so its L2 projection is computable from degrees of freedom while retaining Pk and the same nodal degrees of freedom.The construction also equips the extended space with moments.Source: §2 · p. 3

3.1 The PAMPA scheme

• The local degrees of freedom comprise boundary point values and a polygon average; for k = 2, the nodal values are at vertices and edge midpoints.The basis functions are introduced for discussion, not used in practice.Source: §3.1 · p. 3
• At a vertex, the scaled normal is the sum of the scaled outward normals of its two sharing faces; at a nonvertex edge point, it is the edge’s outward normal.Source: §3.1 · p. 3
• The filtering term removes spurious wiggles that can arise when the projected gradient is zero although the solution is nonconstant on a polygon.Source: §3.1 · p. 4
• The system is evolved in time with a strong-stability-preserving Runge–Kutta scheme.Source: §3.1 · p. 4

3.2 Analysis of the PAMPA scheme

• The dual basis recovers moments as L2 inner products, but its associated mass matrix is positive definite and impossible to compute directly.Source: §3.2 · pp. 4–5
• The PAMPA scheme can be interpreted through a dG semi-discrete formulation, separating updates for the average value and the point values.Source: §3.2 · p. 5
• For linear fluxes with constant A, the analysis begins from the fact that J · ∇u may not belong to V2(K), even for linear problems.The projected flux is approximated using π0 and ∇(π0u), with an O(h^3_K) error under the stated constant-A assumption.Source: §3.2 · pp. 5–6
• Two approximations address this computability issue: approximate the dual basis through a computable VEM mass matrix, or use a cheaper estimate.The authors report using the cheaper option because it is simpler.Source: §3.2 · pp. 6–7

4 Boundary conditions

• To impose a flux boundary condition on a boundary face, add its boundary flux integral to the corresponding dG term.Source: §4 · p. 7
• Only degrees of freedom on the boundary face contribute, because the associated basis functions vanish there for degrees of freedom off that face.Source: §4 · p. 7
• The resulting construction replaces the non-computable inverse mass matrix with the approximation defined earlier.The boundary contribution uses the approximated dual functions.Source: §4 · p. 8
• On simplex meshes, the same technique can use the exactly known mass matrix, and the stabilisation term ΨKσ is not needed.Source: §4 · p. 8

5 Nonlinear stabilization for PAMPA-VEM

• The standard third-order PAMPA method can oscillate near discontinuities, motivating limiting to address potentially negative density or internal energy in gas dynamics.A first-order PAMPA scheme and convex monolithic limiting approach were introduced to ensure positivity.Source: §5 · p. 8
• The polygonal-mesh generalization of the convex oscillation-elimination parameter targets spurious oscillations near shocks and strong discontinuities.The parameter was introduced for unstructured triangle meshes in [5], inspired by oscillation-free and oscillation-elimination DG methods.Source: §5 · p. 8
• The strategies blend a first-order Local Lax-Friedrichs scheme with the high-order scheme; combining them yields a bound-preserving, oscillation-eliminated, entropy-satisfying scheme.Source: §5 · p. 8

6.1 Shock Vortex interaction

• The shock-vortex simulation uses a 180 × 600 quadrilateral mesh with 107,221 elements and 323,220 degrees of freedom.The mesh is treated as polygons; the domain has inflow/outflow sides and top and bottom walls.Source: §6.1 · p. 8
• The shock-vortex example plots density isolines at three simulation times and a sensor whose values range from about 0.75 to 1.The caption assigns red to 1 and yellow to about 0.75.Source: Figure 2 · p. 9

6.2 Shock diffraction over a wedge

• The wedge-diffraction case uses a Mach 10 shock on a mesh of 108,280 polygons and 317,887 degrees of freedom, and runs to T = 8 × 10^-3.The mesh contains quadrilaterals and triangles, both treated as polygons.Source: §6.2 · p. 9
• In the wedge-diffraction example, first-order treatment is confined to the corner and strongest shocks, while the rest of the solution is virtually high order.The reported observation describes where the first-order scheme is applied.Source: §6.2 · pp. 9–10
• The wedge-diffraction figures show a zoomed mesh with density isolines and a sensor, plus density and pressure isolines spanning 0.07584–5.735 and 0.03788–116.9, respectively.The isoline plots use 30 equally spaced contours.Source: Figure 3 · p. 9, Figure 4 · p. 10

• For semi-discrete Active Flux-type PAMPA on polygonal meshes, the paper clarifies boundary-condition implementation and the need for a filtering mechanism.The conclusion identifies both as matters clarified by the work.Source: §7 · p. 10
• The construction extends to higher-order approximation by representing internal moments as L2 moments of polynomial functions, with the average as the moment against the constant function 1.Source: §7 · p. 10