Benchmarking stabilized and self-stabilized p-virtual element methods with variable coefficients

Imagine you are an architect trying to build a bridge across a river. The river isn't straight; it has curves, bends, and the ground beneath it changes from soft sand to hard rock. To build this bridge, you need a blueprint that can handle these irregularities perfectly.

In the world of computer simulations (used to design planes, cars, and bridges), the Virtual Element Method (VEM) is like a super-flexible set of building blocks. Unlike traditional methods that only work with perfect squares or triangles, VEM can use weird, polygon-shaped blocks that fit together perfectly, even if the edges are curved. This is great for modeling complex shapes like the "variable stiffness panels" mentioned in the paper (think of airplane wings where the material gets stiffer in some spots and softer in others).

However, there's a catch. Because these blocks are so flexible, the math inside them gets messy. To make the math work, scientists usually have to add a "stabilizer"—a kind of safety net or glue.

The Problem: The "Glue" Dilemma

The paper starts by pointing out a major headache: How do you choose the right glue?

Standard VEM (The Glue Method): You have to manually pick a "stabilization parameter" (let's call it the Glue Strength).
- If you use too little glue, the bridge collapses (the math becomes unstable).
- If you use too much, the bridge becomes rigid and brittle, or the math gets so messy it takes forever to solve (poor conditioning).
- The tricky part? The "right" amount of glue changes depending on how complex your shape is or how high the "order" (detail level) of your simulation is. It's like trying to guess the exact amount of flour for a cake without a recipe; sometimes it works, sometimes it's a disaster.

The Solution 1: Self-Stabilized VEM (The Self-Healing Bridge)

To fix the glue problem, the authors looked at Self-Stabilized VEM.

The Analogy: Instead of adding external glue, you build the blocks themselves so they are naturally strong. You make the blocks slightly "smarter" by adding extra internal features (degrees of freedom).
The Result: These blocks don't need external glue. They are "self-stabilized."
The Trade-off: While they are very accurate and don't require you to guess a parameter, they are computationally expensive. It's like building a bridge out of solid gold instead of steel; it's incredibly strong and precise, but it's heavy, hard to move, and takes a lot more energy to build. The paper found that while these methods are accurate, they make the computer's math "stiff" (hard to solve) and slow.

The Solution 2: VC-VEM (The Smart Material)

The second half of the paper tackles a specific, difficult scenario: Variable Coefficients.

The Scenario: Imagine your bridge isn't just sitting on different ground; the bridge itself is made of a material that changes its properties as you move along it (e.g., the wood gets harder as you go from left to right).
The Old Way: Standard methods try to approximate this changing material by averaging it out or using simple guesses. For simple shapes, this works. But for high-detail simulations (high "p-order"), this approximation fails, and the bridge starts to wobble.
The New Way (VC-VEM): The authors invented a new way to build the blocks. Instead of just looking at the shape of the block, the new method explicitly looks at the material properties inside the block while it's being built.
The Analogy: It's like a tailor who doesn't just cut a suit based on your height, but also measures your muscle density, bone structure, and posture while cutting the fabric. The result is a suit that fits perfectly, no matter how complex your body is.
The Result: This new method (VC-VEM) is much more robust. Even when the material changes wildly or the simulation gets very detailed, it stays accurate. It doesn't lose its "cool" (accuracy) when things get complicated.

The Big Takeaways

The paper ran thousands of computer tests (benchmarks) comparing these different methods on various shapes (squares, weird polygons, curved edges) and problems (heat flow, stretching metal, fluid flow).

Self-Stabilized methods are great for accuracy but are "heavy" and slow. They are the luxury cars of the simulation world.
Standard Stabilized methods are faster and lighter, but you have to be careful with your "glue" settings. If you get it wrong, the results suffer.
The New VC-VEM is the winner for complex materials. It handles changing materials (like variable stiffness panels) much better than the old ways, especially when you need high precision.

In a Nutshell

The authors are saying: "We tested different ways to make our computer simulations of complex, curvy structures more reliable. We found that while 'self-healing' blocks are accurate, they are slow. But our new 'Smart Material' approach (VC-VEM) is the best way to handle structures where the material properties change from place to place, ensuring your digital bridge doesn't collapse, even when the math gets really hard."

This is a big deal for aerospace engineers who want to design lighter, stronger, and more efficient airplanes with complex, curved parts.

1. Problem Statement

The paper addresses the numerical challenges associated with the Virtual Element Method (VEM), specifically in its p-version (high-order polynomial approximations), when applied to problems with variable coefficients and curvilinear geometries.

Context: The research is motivated by aerospace applications involving variable stiffness panels (e.g., composite structures with curvilinear stringers). These problems require solving partial differential equations (PDEs) with spatially varying elastic properties on complex, non-polygonal meshes.
Core Issue: Standard VEM relies on polynomial projections to handle non-polynomial components of the discrete space. To ensure stability, a stabilization term is added. However, this term is not uniquely defined by the variational formulation and is often chosen ad hoc.
- Limitations: The stabilization parameter can degrade solution quality, cause spurious modes, and lead to suboptimal convergence rates, particularly in high-order ( $p$ -version) approximations and anisotropic problems.
- Variable Coefficients: Existing approaches for handling variable coefficients often rely on constant approximations or low-order polynomial projections, which fail to maintain accuracy at high orders.

2. Methodology

The authors conduct a comprehensive numerical investigation comparing stabilized and self-stabilized VEM formulations across three benchmark problems: the Laplace equation, Linear Elasticity, and the Stokes problem.

A. Stabilized VEM Formulations

The study evaluates several stabilization techniques ( $S_E$ ) used to approximate the non-computable part of the bilinear form:

$S_E^1, S_E^2, S_E^3$ : Standard DOF-based stabilizations (scaling by degrees of freedom or boundary terms).
$S_E^4$ : A scheme specifically designed for $p$ -VEM involving $L^2$ projections on boundaries and interiors.
$S_E^5$ : A matrix-based stabilization effective for linear elasticity.
Parametric Study: The sensitivity of these methods to the stabilization parameter $\tau$ and the rank estimation tolerance (used to determine the necessary polynomial order) is analyzed.

B. Self-Stabilized VEM Formulations

To eliminate the arbitrary stabilization parameter, the authors investigate self-stabilized methods. These approaches enlarge the approximation space or use higher-order projections ( $p^* > p$ ) to ensure the discrete problem is automatically stable. Six strategies are tested:

V1, V4: Enlarged enhanced spaces with $L^2$ or elliptic projections.
V2, V5: Augmented spaces with additional internal degrees of freedom (DOFs).
V3, V6: Divergence-free (or strain-free) spaces using specific projection operators.
Algorithm 1: An iterative procedure is used to determine the minimum additional polynomial order ( $\ell_E$ ) required to ensure the local stiffness matrix is full rank.

C. The VC-VEM Approach (Variable Coefficients)

A novel contribution is the VC-VEM (Variable Coefficient VEM).

Concept: Instead of using standard polynomial projections ( $\Pi^\nabla_p$ or $\Pi^0_{p-1}$ ) that ignore coefficient variability, VC-VEM introduces a new projection operator ( $\Pi^A_p$ or $\Pi^C_p$ ) that explicitly incorporates the variable coefficient matrix ( $A(x,y)$ or $C(x,y)$ ) into the projection definition.
Implementation: The projection is defined via an integration-by-parts formulation that accounts for the spatial derivatives of the coefficient. This allows the method to capture the physics of variable stiffness more accurately within standard VEM spaces.
Applicability: This approach is integrated into both stabilized and self-stabilized frameworks.

3. Key Contributions

Comprehensive Benchmarking: The paper provides the first systematic comparison of stabilized vs. self-stabilized formulations specifically for the p-VEM (high-order) across Laplace, elasticity, and Stokes problems on various mesh types (quadrilateral, Voronoi, octagonal, and Bézier-curved edges).
Analysis of Conditioning: It highlights a critical trade-off: while self-stabilized methods achieve optimal accuracy without tuning parameters, they suffer from worse conditioning (higher condition numbers) and higher computational costs due to the need for higher-order polynomial projections.
New Projection for Variable Coefficients (VC-VEM): The authors propose and validate a new projection operator that explicitly handles variable coefficients. This overcomes the accuracy degradation seen in standard high-order VEM when coefficients vary spatially.
Robustness on Curved Edges: The study demonstrates that the proposed methods perform robustly on meshes with curved edges (Bézier curves), which are essential for modeling curvilinear reinforcements in aerospace structures.

4. Numerical Results

The results were validated on meshes with varying complexity and polynomial orders ( $p=1$ to $12$).

Stabilization Parameter Sensitivity: For stabilized VEM, the choice of $\tau$ significantly impacts accuracy at high orders ( $p > 3$ ). Values that are too small or too large lead to errors or poor conditioning. An optimal range exists, but it requires tuning.
Self-Stabilized Performance:
- Accuracy: Self-stabilized formulations achieve optimal convergence rates comparable to stabilized ones.
- Conditioning: They consistently exhibit worse conditioning than stabilized versions.
- Tolerance: The tolerance used to determine the additional polynomial order ( $\ell_E$ ) affects conditioning but not accuracy. A "looser" tolerance improves conditioning by increasing $\ell_E$ but raises computational cost.
Variable Coefficients (VC-VEM):
- Standard Approaches: Standard stabilized VEM with $\Pi^\nabla_p$ suffers a strong loss of accuracy for $p \ge 3$ with variable coefficients. The $\Pi^0_{p-1}$ approach works well for polynomial coefficients but degrades for trigonometric coefficients at high $p$ .
- VC-VEM Performance: The proposed VC-VEM maintains optimal accuracy for both polynomial and trigonometric variable coefficients, even at very high orders ( $p \ge 10$ ). It is significantly more robust than existing methods.
- Cost: While VC-VEM requires recomputing projections when coefficients change, the cost is mitigated in the p-version context where coarse meshes are typically used.

5. Significance and Conclusion

This paper is significant for the computational mechanics community, particularly in the field of aerospace structural analysis involving composite materials.

Practical Impact: It provides a roadmap for selecting VEM formulations. If computational resources are limited and mesh conditioning is a concern, stabilized VEM with a carefully tuned parameter is sufficient. If maximum accuracy and parameter-free robustness are required (despite higher conditioning), self-stabilized VEM is preferred.
Theoretical Advancement: The introduction of VC-VEM solves a long-standing limitation in high-order VEM regarding variable coefficients. It enables the accurate simulation of complex, variable-stiffness structures on general polytopal meshes with curved boundaries, which was previously difficult with standard high-order methods.
Future Work: The authors suggest extending these formulations to geometrically nonlinear elasticity problems.

In summary, the paper establishes that while self-stabilized methods offer a parameter-free route to accuracy, the VC-VEM approach is the most robust solution for high-order simulations of problems with spatially varying material properties.