🔢 mathematics

A penalized ϕ-FEM scheme for the Poisson Dirichlet problem

This paper presents and analyzes a penalized variant of the $\phi$ -FEM scheme for the Poisson Dirichlet problem that enforces boundary conditions via penalization and ghost penalty stabilization, achieving optimal $H^1$ and quasi-optimal $L^2$ convergence rates while requiring the level-set function only on boundary-adjacent cells.

Original authors: Raphaël Bulle, Michel Duprez, Vanessa Lleras, Killian Vuillemot

Published 2026-02-06

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Raphaël Bulle, Michel Duprez, Vanessa Lleras, Killian Vuillemot

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to paint a picture of a complex shape, like a jagged rock or a liver, on a perfectly square grid of graph paper.

The Old Problem: Cutting the Paper
Traditionally, if you wanted to solve a physics problem (like heat distribution) on that weird shape using the "Finite Element Method" (FEM), you had to cut your graph paper to fit the shape exactly. You'd have to slice the squares along the curve of the rock. This is like trying to fit a square peg in a round hole, but in reverse: you are forcing the grid to bend and twist to match the object. It's accurate, but it's a nightmare to do, especially if the shape is moving or very complicated.

The First New Idea: The "Ghost" Method (Original $\phi$ -FEM)
A few years ago, researchers invented a clever trick called $\phi$ -FEM. Instead of cutting the paper, they kept the perfect square grid. They defined the shape using a "level-set" function (think of it as a topographic map where the sea level is the edge of your rock).

In the original version, they treated the solution (the paint) as if it were a product of two things: the shape itself and a hidden helper variable. It worked great, but it required knowing very specific details about the "terrain" (the level-set function) everywhere on the grid, even far away from the rock. It was like needing a detailed map of the entire ocean just to paint a single island.

The New Idea: The "Penalized" Method
This paper introduces a new, improved version of that trick. Let's call it the "Penalized $\phi$ -FEM."

Here is how it works, using a simple analogy:

The Setup: You still have your square grid and your rock shape. You don't cut the grid.
The "Ghost" Variable: Inside the grid cells that touch the edge of the rock, the authors introduce a new "helper" variable (let's call it $p$ ). They say, "The solution $u$ is just the shape function $\phi$ multiplied by this helper $p$ ."
The Penalty (The "Fine"): In the old method, they forced this relationship to be perfect everywhere. In this new method, they only encourage it to be true near the edge. If the solution and the helper don't match up perfectly near the boundary, the computer adds a "penalty" (a fine) to the math equation.
- Analogy: Imagine you are trying to keep a dog on a leash. The old method was like tying the dog to a post everywhere it went. The new method is like saying, "You can run around, but if you get too far from the edge of the yard, I'll charge you a fine." This keeps the dog (the solution) in line without needing to tie it down everywhere.

Why is this better?

Less Work: You only need to know the details of the "terrain" (the level-set function) right next to the rock's edge. You don't need to know it for the whole grid. This makes the method more robust and less likely to crash if the map is a little fuzzy.
No Singularity Issues: Sometimes, the math for the "terrain" gets weird or breaks down (like at the very center of a circle). Because this new method only looks at the edge, it avoids those tricky spots.
Speed: The authors tested this on a computer. They found that while the math system is slightly bigger (because of the extra helper variable), the method is still much faster than the old "cutting the grid" method. It's almost as fast as the original $\phi$ -FEM but easier to handle.

The Results
The authors did the heavy math to prove that this method is accurate. They showed that as you make the grid squares smaller (more detailed), the answer gets closer to the true solution at the best possible speed.

They tested it on:

A 2D shape that looks like a liver (made of 5 overlapping blobs).
A 3D sphere.

In both cases, their new "Penalized" method performed just as well as the original $\phi$ -FEM and the standard "cut-grid" method, but with the added benefit of being simpler to implement near the boundaries.

In Summary
This paper presents a smarter way to solve physics problems on complex shapes without having to cut your computer grid to fit them. By using a "penalty" system to enforce the rules only at the edges, they created a method that is robust, accurate, and computationally efficient, making it easier to simulate complex real-world objects like organs or mechanical parts.

Technical Summary: A Penalized $\phi$ -FEM Scheme for the Poisson Dirichlet Problem

Problem Statement
The paper addresses the numerical solution of the Poisson equation with homogeneous Dirichlet boundary conditions on complex or evolving geometries. The primary challenge in such problems is the generation of boundary-fitted meshes, which can be restrictive. The authors focus on the Poisson problem:
$\begin{cases} -\Delta u = f & \text{in } \Omega, \\ u = 0 & \text{on } \Gamma, \end{cases}$
where the domain $\Omega$ and its boundary $\Gamma$ are described implicitly by a level-set function $\phi$ such that $\Omega = \{\phi < 0\}$ and $\Gamma = \{\phi = 0\}$ .

Methodology
The authors propose a penalized variant of the $\phi$ -FEM (phi-Finite Element Method), an unfitted finite element method that avoids boundary-fitted meshes. Unlike the original $\phi$ -FEM formulation, which redefines the unknown over the entire domain via $u = \phi w$ , the proposed scheme enforces boundary conditions weakly through a penalization term.

Key methodological features include:

Local Level-Set Requirement: The method requires the level-set function $\phi$ only on the cells adjacent to the boundary ( $\Omega_h^\Gamma$ ), rather than globally. This is achieved by introducing an auxiliary variable $p$ defined on the boundary-adjacent cells such that $u = \phi p$ is enforced via penalization.
Variational Formulation: The scheme seeks $(u_h, p_h) \in V_h^{(k)} \times Q_h^{(k)}$ $(u_{h}, p_{h}) \in V_{h}^{(k)} \times Q_{h}^{(k)}$ satisfying a variational equation that includes:
- Standard gradient terms over the background mesh $\Omega_h$ .
- A boundary integral term.
- A penalization term $\frac{\gamma}{h^2} \int_{\Omega_h^\Gamma} (u_h - \frac{1}{h}\phi_h p_h)(v_h - \frac{1}{h}\phi_h q_h)$ to enforce the relation $u \approx \phi p$ .
- A "Ghost Penalty" stabilization term ( $G_h^{lhs}$ ) involving jumps of normal derivatives and Laplacians on the boundary-adjacent facets and cells to ensure coercivity.
Stabilization: The scheme utilizes a ghost penalty technique with parameters $\gamma$ and $\sigma_D$ to stabilize the system, particularly for elements cut by the boundary.

Key Contributions and Theoretical Results
The paper provides a rigorous numerical analysis of this penalized scheme, which was previously introduced but not analyzed. The main theoretical contributions are:

Coercivity: The authors prove that the associated bilinear form is coercive under suitable assumptions on the mesh and the level-set function, provided the stabilization parameters $\gamma$ and $\sigma_D$ are sufficiently large.
A Priori Error Estimates:
- $H^1$ Semi-norm: The scheme achieves optimal convergence of order $O(h^k)$ , where $k$ is the polynomial degree of the finite element space.
- $L^2$ Norm: The analysis yields a quasi-optimal convergence rate of $O(h^{k+1/2})$ .
- Regularity: The analysis requires less regularity on the solution $u$ compared to the original $\phi$ -FEM analysis found in prior literature.
Conditioning: Theoretical results and numerical experiments suggest the condition number of the associated matrix scales as $O(h^{-2})$ .

Numerical Results
The theoretical findings are validated through numerical experiments implemented in Python using the Dolfinx library:

2D Complex Geometry: A liver-like geometry was tested. The penalized scheme ("Dual $\phi$ -FEM") demonstrated optimal convergence rates in both $L^2$ and $H^1$ norms, comparable to the standard fitted FEM and the original "Direct $\phi$ -FEM."
3D Geometry: A spherical domain was tested with a manufactured solution. Again, optimal convergence was observed.
Performance Comparison: While both $\phi$ -FEM variants were faster than standard fitted FEM (due to avoiding complex mesh generation), the penalized scheme incurred higher computational costs than the direct $\phi$ -FEM due to the increased system size from the additional variable $p$ . However, the penalized scheme offers robustness regarding the level-set function, as it does not require gradients or second derivatives of $\phi$ globally.

Significance and Claims
The paper claims that this penalized approach successfully combines the robustness of mixed formulations with the flexibility of immersed finite element methods. Its primary significance lies in:

Reduced Level-Set Dependency: By restricting the requirement for the level-set function (and its derivatives) to the boundary-adjacent cells, the method is more robust to singularities in the level-set function (e.g., in signed distance functions) and does not require global reconstruction of the solution unknown.
Compatibility: The local nature of the level-set requirement makes the scheme compatible with Neumann boundary conditions, as presented in related works.
Validation: The work provides the first rigorous error analysis for this specific penalized variant, confirming its theoretical convergence properties and practical viability for complex geometries.

The authors note that while the theoretical $L^2$ estimate is quasi-optimal, numerical experiments in the final section suggest optimal convergence of order $O(h^{k+1})$ in practice.

Technical Summary: A Penalized ϕ\phiϕ-FEM Scheme for the Poisson Dirichlet Problem

More like this

Technical Summary: A Penalized $\phi$ -FEM Scheme for the Poisson Dirichlet Problem