Orthogonal Null-space Eliminated EARM

• ON-EARM is a robust framework for flux recovery in conforming finite element methods that eliminates the divergence-free null space via orthogonal projection.
• It combines a weighted averaging step with a unique residual correction to construct locally conservative H(div)-conforming fluxes while addressing nonuniqueness.
• Numerical validations on benchmark problems demonstrate that ON-EARM supports arbitrary polynomial orders and achieves optimal convergence with efficient a posteriori error estimators.

The Orthogonal Null-space–Eliminated Equilibrated Averaging Residual Method (ON-EARM) is a robust and efficient framework for flux recovery in conforming finite element methods for elliptic interface problems. ON-EARM is developed as a conforming variant within the broader Equilibrated Averaging Residual Method (EARM) framework, specifically to address nonuniqueness in the flux correction step for conforming discretizations by eliminating the divergence-free null space through orthogonal projection. This methodology is applicable in both two and three dimensions, supports arbitrary polynomial order, yields fluxes locally conservative with respect to the strong PDE, and enables construction of rigorous a posteriori error estimators with guaranteed reliability and efficiency, independent of coefficient jumps or mesh grading (He, 4 Jan 2026, He, 5 Mar 2025).

1. Mathematical Framework and Model Problem

ON-EARM operates in the context of the discontinuous-coefficient diffusion equation: $$-∇·(A ∇u)=f \text{ in } Ω, \quad u=0 \text{ on } Γ_D, \quad -A∇u·n=g \text{ on } Γ_N,$$ where Ω ⊂ ℝ^d is a bounded Lipschitz domain, A(x) is symmetric, piecewise-constant, positive-definite, and f, g denote source and Neumann data. The finite element mesh 𝒯_h consists of elements K, with facets ℰ partitioned as interior (ℰ_I), Dirichlet (ℰ_D), and Neumann (ℰ_N).

Finite element solutions are considered in the spaces CG_{0,Γ_D}(𝒯_h, k) (conforming) and the Raviart–Thomas space RT(𝒯_h, s) for flux recovery, with polynomial degree parameter s (typically s = k–1 for conforming FEM). The goal is to recover a flux $\hatσ_h ∈ H(\text{div}; Ω)$ that is locally conservative: ∇·\hatσ_h = f in each element and $\hatσ_h·n = g$ on Neumann boundaries.

2. The EARM Approach and Its Difficulties in the Conforming Case

EARM proceeds in two steps:

1. Averaging step: Construct an H(div)-conforming field $\tildeσ_s$ by weighted averaging of the broken numerical flux −A∇u_h, matching prescribed facet and cell moments.
2. Residual correction: Compute a correction σ_s^Δ ∈ RT(𝒯_h, s) to enforce strict local conservation (div ( $\tildeσ_s + σ_s^Δ$) = f).

For discontinuous Galerkin (DG) and nonconforming methods, the correction problem is explicit or locally posed. In the conforming FEM case, however, the correction equation

$$(\text{div } σ^Δ, v) = r(v) := (f-∇·\tildeσ_{k-1}, v)$$

for v in DG(𝒯_h, k–1) is not well-posed: the kernel—comprising divergence-free fluxes orthogonal to prescribed moments—renders the solution nonunique (He, 4 Jan 2026, He, 5 Mar 2025).

3. Null-space Structure and the ON-EARM Solution

The null space N is characterized as

$$N = \{ \tau ∈ RT(𝒯_h, s) : ∇·τ = 0,\, \tau·n|_{ℰ_N}=0,\, (\tau, ψ)_K=0\, ∀ψ ∈ P_{s-1}(K)^d \},$$

which is infinite-dimensional and consists of fluxes that do not affect the divergence or Neumann flux conditions. To enforce uniqueness, ON-EARM restricts the flux correction to the orthogonal complement N^⊥ with respect to a bilinear form defined on facet jumps: $$𝒜(w,v) = \sum_{F∈ℰ\setminusℰ_N}\int_F A_F h_F^{-1}[w][v]\,ds,$$ where [w] represent facet jumps and A_F, h_F denote the diffusion coefficient and facet diameter, respectively.

A surjective lifting operator S: DG(𝒯_h, s) → RT(𝒯_h, s) maps discrete facet functions to Raviart–Thomas fluxes, with

$S(w)·n_F = A_F h_F^{-1}[w].$

One then seeks a correction u_s^Δ ∈ DG⁰(𝒯_h, s) (functions modulo continuous subspace) such that

$$𝒜(u_s^Δ, v) = R(v) = (f − ∇·\tildeσ_{k-1}, v)_Ω \qquad ∀ v ∈ DG⁰(𝒯_h, s).$$

This global but low-dimensional linear system is coercive and thus admits a unique solution by Lax–Milgram.

4. Construction of the ON-EARM Recovered Flux

With the unique u_s^Δ identified, ON-EARM defines the recovered flux as

$$\hatσ_h^{cg} = \tildeσ_{k-1}^{cg} + S(u_{k-1}^Δ) ∈ RT_f(𝒯_h, k-1, k-1),$$

or for lower order, as $\tildeσ_s^{cg} + S(u_s^Δ)$. In both cases, the reconstructed flux is exactly equilibrated (locally conservative) and conforms to the H(div) requirement, ensuring compatibility with physical conservation laws and boundary fluxes.

This orthogonal projection mechanism distinguishes ON-EARM from classical equilibrated residual methods, which lack either uniqueness or robustness with respect to coefficient jumps in A.

5. A Posteriori Error Estimation and Theoretical Properties

ON-EARM yields a local a posteriori error indicator on each element K defined by

$$η_{σ,K} = \|A^{-1/2}(\hatσ_h - (–A∇u_h))\|_{L²(K)},$$

and a global estimator

$$η_σ = (\sum_K η_{σ,K}^2)^{1/2} = \|A^{-1/2}(\hatσ_h + A∇u_h)\|_{L²(Ω)}.$$

The theoretical foundation relies on a generalized Prager–Synge identity: $$‖A^{1/2}∇(u–w)‖^2 = \inf_{τ∈Σ_f(Ω)}‖A^{-1/2}τ + A^{1/2}∇_h w‖^2 + \inf_{v∈H¹_D}‖A^{1/2}∇_h(v–w)‖^2.$$ For conforming solutions, the nonconforming term vanishes, and the estimator bounds the true error robustly: $$‖A^{1/2}∇(u–u_h^{cg})‖ ≤ η_σ + C\,\operatorname{osc}(f).$$ Efficiency (lower bounds) holds with constants independent of mesh size and jumps in A under local quasi-monotonicity. These properties guarantee that the error estimator provides both rigorous upper and lower bounds for the energy norm error, with effectivity indices observed in practice uniformly bounded (∼1–7), regardless of problem singularities or coefficient variations (He, 4 Jan 2026, He, 5 Mar 2025, He et al., 2020).

6. Algorithmic Implementation

The ON-EARM workflow involves the following steps:

1. Averaging: Assemble $\tildeσ_s$ by matching facet averages of –A∇u_h and cell moments.
2. Residual: Compute the averaging residual r_s(v) = (f – ∇·\tildeσ_s, v).
3. Correction: Solve the global symmetric system 𝒜(u_s^Δ, v) = R(v) for u_s^Δ on facet-DOF quotient space.
4. Recovery: Set $\hatσ_h := \tildeσ_s + S(u_s^Δ)$.
5. Error Estimation: Calculate local and global estimators η{σ,K}, ησ as detailed above.

The linear system in Step 3 is of size ≃ number of facets × polynomial order, significantly smaller than the global PDE system. In two dimensions, further localization is possible using Gauss–Lobatto quadrature.

7. Numerical Validation and Practical Performance

ON-EARM has been validated on benchmark problems including the Kellogg interface (large jumps in A), L-shaped domains (corner singularities), and Fichera corners in 3D. The estimator was employed in adaptive refinement cycles (SOLVE→ESTIMATE→MARK→REFINE), achieving optimal convergence rates of ‖∇(u–u_h)‖ ∼ N^{–k/d} for all polynomial degrees k = 1, 2, 3. Effectivity indices remained uniformly bounded and robust against both coefficient jumps and solution singularities. Superconvergence was observed in some cases for k ≥ 2 (He, 4 Jan 2026).

---

References:

He & Cuiyu et al., "A Unified Equilibrated Flux Recovery Framework with Robust A Posteriori Error Estimation" (He, 4 Jan 2026). He, Cai & Zhang, "Equilibrated Averaging Residual Method: A General Approach to Conservative Flux Recovery" (He, 5 Mar 2025). Cai, He & Zhang, "Generalized Prager-Synge Inequality and Equilibrated Error Estimators for Discontinuous Elements" (He et al., 2020).