Defect Subspaces and Localized Instabilities in Cut-Cell Finite-Volume Operators

Technical Summary: Defect Subspaces and Localized Instabilities in Cut-Cell Finite-Volume Operators

Problem Statement
Explicit finite-volume methods for embedded boundary (cut-cell) problems suffer from a persistent stability issue known as the "small cut-cell" problem. When an embedded boundary clips a background cell to a small volume fraction $\alpha \ll 1$, the local Courant-Friedrichs-Lewy (CFL) number becomes $\lambda_\alpha = \lambda/\alpha \gg 1$, even if the background CFL $\lambda$ is stable. This forces the local update operator to possess large unstable eigenvalues, often destroying the computation in a single time step. While various stabilization strategies exist—such as flux redistribution, cell merging, and state redistribution (SRD)—the underlying spectral mechanism of this instability has not been fully characterized. Prior work, such as [2], established stability thresholds (e.g., $\lambda_\alpha \le 2$ for 1D upwind) but did not identify the dominant eigenvalues, the localization of eigenvectors, or the specific defect matrix governing the growth rate. Consequently, practitioners rely on empirical parameter tuning or conservative global bounds, leading to inefficient simulations or costly crashes in production codes.

Methodology
The paper employs a rigorous spectral analysis of linearized explicit finite-volume operators on periodic cut-cell meshes. The approach proceeds through a logical chain: small-cell geometry $\to$ defect matrix $\Gamma$ $\to$ unstable spectrum $\to$ stabilization requirements.

1. One-Cut-Cell Analysis (1D): The author derives an explicit characteristic polynomial for a periodic upwind operator with a single cut cell. Using Rouché's theorem and the contraction mapping theorem, they isolate a unique dominant eigenvalue $\mu_u$ and prove that the corresponding eigenvector $v_u$ is exponentially localized at the cut cell ($v_u = e_c + O(\alpha)$).
2. Singular-Block Framework: The paper abstracts the cut-cell operator into a block matrix form $A_\alpha = \begin{pmatrix} -\alpha^{-p}\Gamma + R_\alpha & B_\alpha \\ P_\alpha & Q_\alpha \end{pmatrix}$. Here, $\Gamma$ is the "defect matrix" encoding the local geometry, and $p$ is a scaling exponent. Theoretical results (Theorems 6.2 and 6.3) prove that this structure yields exactly $m$ unstable eigenvalues scaling as $O(\alpha^{-p})$ and that the unstable spectral subspace converges to the defect coordinate subspace at a rate of $O(\alpha^p)$.
3. Operator Characterization: Section 7 identifies the precise conditions under which finite-volume operators admit this singular-block structure, specifically requiring that flux coefficients scale as power laws of $\alpha$. This covers schemes from first-order Godunov to high-order MUSCL with bounded flux coefficients.
4. Stabilization Analysis: The paper analyzes the "merge-left" State Redistribution (SRD) method. It proves that the SRD correction vector is asymptotically aligned with the unstable eigenvector ($|\cos_\alpha(D, v_u)| = 1 - O(\alpha)$), explaining why SRD effectively damps the instability.
5. Geometry-Based Criterion: By analyzing the defect matrix $\Gamma$ for specific schemes, the author derives a deterministic formula for the minimum SRD blending parameter $s_{stab}$ required to stabilize a cut cell, computable solely from local mesh geometry ($\alpha$ and $\lambda$) without global matrix assembly or time-stepping.

Key Contributions and Results

• Spectral Characterization: The paper proves that for $m$ small cut cells, the update operator has exactly $m$ unstable eigenvalues, each localized at a specific cut cell. The dominant eigenvalue is $\mu_u = 1 - \lambda/\alpha + O(1)$, and the eigenvector is $v_u = e_c + O(\alpha)$.
• Defect Matrix $\Gamma$: The growth rate of the instability is governed by a local defect matrix $\Gamma$. For scalar 1D schemes with bounded flux coefficients, $\Gamma = (\Delta t/h)(a_{c+1/2,c} - a_{c-1/2,c})$, where $a$ are the linearized flux coefficients at the cut-cell faces.
• SRD Alignment Theorem: It is proven that SRD acts in the unstable eigendirection. The correction vector $D$ satisfies $|\cos_\alpha(D, v_u)| = 1 - O(\alpha)$, providing a theoretical justification for SRD's efficacy without needing scheme-specific stability proofs.
• Stability Criterion: The paper derives a geometry-based stability criterion. A cut cell is unstable if $\lambda_\alpha > 2$. To stabilize it, the SRD blending parameter $s$ must satisfy:
  $$s \ge s_{stab} = \frac{\lambda_\alpha - 2}{\lambda_\alpha - 1} + O(\alpha)$$
  This formula allows the required stabilization strength to be computed at mesh generation time for each cut cell individually.
• Multi-Cut-Cell Extension: For $m$ non-adjacent cut cells (separation $d_{min} \ge 2$), the paper proves that the unstable spectral projector converges to the defect subspace with error $O(\alpha)$, provided the cut cells are sufficiently separated.
• Numerical Verification: Extensive numerical experiments (Experiments 1–9) verify the theoretical predictions. These include the exponential convergence of the eigenvalue error with grid size, the $O(\alpha)$ localization of eigenvectors, the sharp instability threshold at $\lambda_\alpha = 2$, and the accuracy of the $s_{stab}$ formula. The defect matrix formula is shown to predict the dominant eigenvalue to machine precision across varying wave speeds and cut-cell positions.

Significance and Claims
The paper claims to provide the first unified algebraic explanation for cut-cell instability, moving beyond empirical observation to a deterministic framework. Its primary significance lies in transforming stability from a trial-and-error process into a checkable property of the mesh geometry.

• Predictive Capability: The derived criterion allows practitioners to flag unstable cut cells and compute the exact minimum blending parameter required before running a simulation. This eliminates the need for rerunning production simulations due to crashes caused by poorly stabilized cut cells.
• Certification of Methods: The paper establishes that any stabilization method can be certified for cut-cell stability by verifying a single algebraic alignment condition ($|\cos_\alpha(D, v_u)| = 1 - O(\alpha)$), rather than requiring a new, complex stability proof for each specific scheme.
• Cost Efficiency: The stability check requires only $O(m)$ operations at mesh generation time, independent of the total mesh size or number of time steps. This contrasts with the high cost of simulation failures, which the paper notes can waste thousands of dollars in cloud HPC resources.
• Scope: The results are rigorously proven for scalar, linear, periodic, one-dimensional problems and extended to $d$-dimensions for a single cut cell. The paper notes that extensions to nonlinear systems, adjacent cut cells ($d_{min}=1$), and non-periodic boundaries remain open problems.

In summary, the paper argues that the "small cell problem" is not an intractable numerical artifact but a localized spectral phenomenon governed by local geometry. By characterizing the defect subspace and the defect matrix, the author provides a practical, assembly-free tool to ensure stability in cut-cell methods.