Subgrid Face Value Reconstruction

• The paper introduces techniques that recover finely-resolved face values from coarse mesh averages using semi-Lagrangian and WENO-like local reconstruction strategies.
• It achieves high-order accuracy and stability in challenging regimes, such as advection-dominated flows and shallow water equations with partial wet/dry conditions.
• The approach reduces computational cost by substituting full fine-mesh evaluations with localized subgrid reconstructions that ensure well-balanced fluxes and consistent source term computations.

Subgrid face value reconstruction refers to numerical strategies for obtaining representative solution values on mesh faces or edges at a subgrid level within multiscale discretizations. This process is critical in finite volume, finite element, and multiscale approaches where enhanced resolution of localized physical features is required, but the primary variables reside exclusively on a coarse mesh. Recent research details procedures for reconstructing physically consistent face values in challenging regimes, including advection-dominated multiscale flows and shallow water equations with fine-scale bathymetric features, ensuring well-balanced fluxes, accuracy, and robust handling of degenerate (partially wet or dry) states (Simon et al., 2019, Bitsch et al., 5 Jan 2026).

1. Mathematical Context and Motivation

Subgrid face value reconstruction arises in numerical schemes where the global system operates on a coarse mesh—often dictated by computational efficiency—while subgrid (finer mesh) features are resolved to capture critical small-scale phenomena. Examples include advection–diffusion problems with highly oscillatory coefficients (Simon et al., 2019) and shallow water equations with intricate bathymetry (Bitsch et al., 5 Jan 2026). In these frameworks, only cell averages or nodal values of key variables are stored on coarse cells, necessitating sophisticated reconstruction procedures to achieve accurate local face values for flux and source term discretization. A typical setup involves:

• Coarse triangulation $T_H$ or $\tilde{T}_m$ with mesh-size $H$.
• Each coarse cell contains a fine subgrid mesh $T_h^K$ or $T_{\ell(m,k)}$ with mesh-size $h\ll H$.
• Solution variables (such as water depth, discharge, velocity) are averaged over coarse cells, while bathymetric and related features are resolved on the subgrid.

2. Reconstruction Methodologies

Semi-Lagrangian Multiscale Basis Construction

For advection-dominated problems, a semi-Lagrangian procedure is used to generate subgrid-corrected basis functions. In each time step, nodal solution values are traced backward along characteristics to define a departure cell $\widetilde K$ (Simon et al., 2019). Local basis functions $\widetilde\phi_{K,j}$ are computed by solving a Tikhonov-regularized inverse problem:

$$\min_{\{\widetilde\phi_{K,j}\}} \Bigl\|u^n(x) - \sum_{j=1}^{N_K}u_j^n\,\widetilde\phi_{K,j}(x)\Bigr\|_{L^2(\widetilde K)}^2 + \sum_{j=1}^{N_K}\alpha_j\,\mathcal{R}_j(\widetilde\phi_{K,j})$$

subject to interpolation constraints at traced-back nodes. These basis functions allow for a local reconstruction operator $R_h^K$ mapping coarse nodal values $\{u_j^n\}$ onto a fine-scale subgrid representation $u_h^n(x)$.

Unstructured Subgrid Procedures for Shallow Water Equations

In unstructured finite volume schemes, reconstructed face values are required for each fine edge covering a coarse cell face. The process involves:

• WENO-like least-squares reconstructions to produce linear polynomials for free-surface elevation and velocities on each coarse cell.
• Evaluation of these polynomials at subgrid edge midpoints to assemble local states $(d, h, \eta, u, v)$.
• Identification of local wet/dry status: a cell is "wet" if $h>0$, otherwise "dry."

Distinct cases are handled depending on the wet/dry status of adjacent subcells:

Case	Description	Formula for $(d_L, h_L, \eta_L)$ and $(d_R, h_R, \eta_R)$
1.1	Both sides wet, no spurious dry patch	$d_{L/R}=(d_l+d_r)/2$; $h_{L/R}=\max(\eta_{l/r}+d_{L/R},0)$; $\eta_{L/R}=h_{L/R}-d_{L/R}$
1.2	Partial dryness	$d_{L/R}=-\min(-\min(d_l,d_r),\min(\eta_l,\eta_r))$; $h_{L/R}=\min(\eta_{l/r}+d_{L/R},h_{l/r})$; $\eta_{L/R}=h_{L/R}-d_{L/R}$
2.1	One side wet, other side dry, no transport	$d_{L}=d_l$, $h_{L}=\eta_l+d_l$; $d_{R}=\bar d_{m_i}$, $h_{R}=\bar \eta_{m_i}+\bar d_{m_i}$
3	Both sides dry (wall flux)	$d_{L}=\bar d_m$, $h_{L}=\bar \eta_m+\bar d_m$; $d_{R}=\bar d_{m_i}$, $h_{R}=\bar \eta_{m_i}+\bar d_{m_i}$

This structure ensures local consistency in the presence of partial flooding, dry interfaces, and discontinuous bathymetry (Bitsch et al., 5 Jan 2026).

3. Flux and Source Term Computation

Once reconstructed face values are available, fluxes across coarse face boundaries are assembled. For advection-diffusion, left/right face trace values $\{u_h^-, u_h^+\}$ are expressed as linear combinations of multiscale basis functions, and upwind or average projections are utilized for flux evaluation:

$$\bar u_F = \frac{1}{|F|} \int_F u_h^\mathrm{upw}(x)\,ds$$

where $u_h^{\mathrm{upw}}(x)$ is selected by the sign of $\beta\cdot n_F$ (Simon et al., 2019). For shallow water systems, a Riemann solver (HLLC) is invoked when there is transport or a wall-flux for impermeable boundaries; the corresponding “star state” (depth, free surface, bathymetry) is combined with edge lengths and normals to assemble the total coarse cell flux.

Gravity source terms are treated via:

$$S_g(U_m) = \frac{1}{|\tilde T_m|} \frac{1}{2}g \sum_{i=1}^{3} \sum_{(k,j)\in\mathcal{E}_{m,i}} (η^*_{kj} + \bar η_m)(d^*_{kj} - \bar d_m) |E_{kj}| n_{m,i}$$

This approach guarantees exact well-balance in lake-at-rest states, even with partial wet cells, eliminating spurious flows (Bitsch et al., 5 Jan 2026).

4. Accuracy, Stability, and Robustness

Numerical tests indicate that subgrid face value reconstruction yields high-order accuracy and robust stability in scenarios that defeat traditional coarse mesh schemes. For advection-dominated problems with unresolved oscillatory coefficients ($H \gg \varepsilon,\delta$), semi-Lagrangian multiscale reconstruction achieves $L^2$ errors of order $10^{-2}$–$10^{-3}$, compared to $10^{-1}$ for standard FEM, and recovers classical convergence rates when sufficiently resolved (Simon et al., 2019). In shallow water schemes, second-order spatial accuracy is achieved through linear WENO reconstructions, and time integration via second-order TVD Runge-Kutta leads to formal second-order accuracy (Bitsch et al., 5 Jan 2026).

In partially wet regions, the algorithms enforce positivity by clamping $h = \max(0, \dots)$ and restricting WENO stencils to all-wet subcells. When only dry stencils are available, piecewise-constant reconstructions ensure first-order accuracy and avoid nonphysical solutions.

5. Extensions, Parallelizability, and Computational Implications

Both the semi-Lagrangian multiscale frameworks and subgrid shallow water methods are well-suited for parallel computation. The reconstruction and local face flux calculations are inherently cell-local, enabling efficient assembly for large-scale simulations. Potential nonlinear extensions include replacing linear inverse problems with PDE-constrained optimization to address reaction–diffusion or nonlinear advection, embedding the methodology within discontinuous Galerkin frameworks, and adapting the reconstruction for solution-dependent coefficients or direct data-assimilation (Simon et al., 2019).

The reduction in required Riemann solves—from fine mesh complexity to coarse face sampling—offers substantial computational savings while maintaining accuracy and resolution of small-scale physical effects (Bitsch et al., 5 Jan 2026). The approach facilitates stable handling of moving flood–dry boundaries and is beneficial when increased subgrid resolution is required to resolve critical features without a full fine-mesh computation.

6. Common Misconceptions and Objective Assessment

A common misconception is that subgrid face value reconstruction entails solving for velocities or fluxes directly on the fine mesh; in reality, only bathymetric and related data are resolved at this level—all flux and source-term computations are carried out from reconstructed local states sampled at fine-edge midpoints. Classical fine-mesh schemes may fail to capture well-balanced behavior or produce spurious flow in lake-at-rest or partial wetness states, but subgrid reconstruction methods specifically address and prevent these issues through local consistency formulas and careful source term balancing (Bitsch et al., 5 Jan 2026). A plausible implication is that, for highly multiscale advection or variable bathymetry, employing subgrid face value reconstruction on a coarse mesh with localized refinement is computationally efficient and physically accurate compared to uniform fine-mesh solutions.