Geometric Reinitialization for Capillary Flows: a Comparative Study with State-of-the-Art Conservative Level-Set Methods

This paper presents a novel geometric reinitialization method for the Conservative Level-Set solver in capillary flow simulations, demonstrating through comparative 3D case studies that it achieves high-fidelity results with greater robustness and fewer parameters than traditional PDE-based approaches, while outperforming simple projection-based methods.

The Gist

Imagine you are trying to simulate how a drop of oil moves through water, or how a bubble rises in a glass of soda. In the world of computer simulations, this is tricky because the boundary between the two fluids (the "interface") is constantly stretching, squishing, and changing shape.

To track this invisible boundary, computers use a mathematical "map" called a Level-Set. Think of this map like a topographic map where the "sea level" (the zero line) represents the exact edge of the drop or bubble.

The Problem: The Map Gets Blurry

As the simulation runs, the computer's math naturally causes this map to get "fuzzy" or "blurry" over time, like a watercolor painting left in the rain. The sharp edge of the drop smears out. If the edge gets too blurry, the computer loses track of how much liquid is actually there (volume loss) and gets the physics of surface tension (the "skin" holding the drop together) wrong.

To fix this, scientists use a process called Reinitialization. This is like taking a blurry photo and running it through a sharpening filter to make the edges crisp again.

The Study: Three Ways to Sharpen the Image

The authors of this paper tested three different "sharpening filters" to see which one works best for complex 3D fluid flows:

1. The "PDE" Method (The Complex Recipe):

   • How it works: This is the current industry standard. It uses a complex set of mathematical rules (equations) to push the blurry edges back into a sharp line.
   • The Catch: It's like trying to bake a perfect cake using a recipe with four different knobs to adjust (temperature, time, mixing speed, etc.). You have to tweak these knobs differently for every single type of cake (or fluid flow) you make. If you get the settings wrong, the cake falls flat.
   • Result: It works very well and gives accurate results, but it is finicky and requires a lot of manual tuning.

2. The "Projection" Method (The Quick Fix):

   • How it works: This is the simplest approach. It just forces the numbers to be sharper instantly, like squishing a sponge back into shape.
   • The Catch: It's too blunt. The paper found that for 3D flows, this method is like trying to fix a broken vase with duct tape—it fails to capture the complex movements. The drop or bubble often disappears or moves to the wrong place.
   • Result: It failed in the 3D tests.

3. The "Geometric" Method (The New Tool):

   • How it works: This is the new method proposed by the authors. Instead of solving complex equations to fix the blur, it uses pure geometry. It literally measures the distance from the edge of the drop to every point in the space around it, rebuilding the map from scratch based on shape.
   • The Benefit: It only requires two knobs to adjust, and those settings work perfectly for every type of flow they tested. It's like having a universal remote that works on every TV brand without needing to change batteries or codes.
   • Result: It produced high-quality, accurate results just as good as the complex method, but it was much more robust and easier to use.

The Tests: Putting Them to the Test

The team tested these methods on three specific scenarios:

• The Rising Bubble: A bubble floating up through a liquid.
• The Migrating Droplet: A drop moving because of a chemical "wind" (surface tension gradient).
• The Breaking Jet: A stream of liquid that breaks apart into droplets (like water from a faucet).

The Findings:

• The Geometric and PDE methods both did a great job. They kept the volume of the drops accurate and showed the correct shapes.
• The Projection method failed miserably in 3D, losing the shape of the drops and getting the physics wrong.
• The Geometric method was the winner because it didn't need constant tweaking. The PDE method worked well but required the user to be a "tuning expert" for every new problem.

The Bottom Line

If you want to simulate how fluids behave in 3D, you need a way to keep the edges of your simulation sharp. This paper shows that a new geometric approach is a "set-it-and-forget-it" solution that is just as accurate as the current complex standard, but much easier to use because it doesn't require constant, case-by-case adjustments. It's a more reliable tool for the computer scientist's toolbox.

Technical Summary

Technical Summary: Geometric Reinitialization for Capillary Flows

Problem Statement
Simulations of immiscible flows driven by surface tension (ST), such as those involving droplets and bubbles, require robust high-fidelity frameworks. While Eulerian multi-phase models like the Conservative Level-Set (CLS) approach are effective, they suffer from interface smearing and volume loss over time due to numerical diffusion. To maintain a sharp interface and ensure volume conservation, reinitialization methods are necessary. However, existing reinitialization strategies—ranging from PDE-based solvers to projection techniques—often involve complex parameter selection, case-dependent tuning, or fail to capture complex 3D interfacial dynamics accurately. There is a lack of objective comparative studies between these methods within a unified numerical framework.

Methodology
The authors present a complete description of a CLS solver implemented within the open-source finite element framework Lethe (based on deal.II). The solver utilizes a one-fluid formulation with Eulerian interface representation, employing Lagrange tensor product elements ($Q_1$) and dynamic mesh adaptation to mitigate numerical diffusion.

The core of the study is a comparative analysis of three distinct reinitialization methods applied to three 3D benchmark cases:

1. PDE-based Reinitialization: Based on the original CLS method by Olsson et al., this approach solves an artificial time-dependent PDE to balance compressive and diffusive terms, restoring the phase indicator to a hyperbolic tangent profile. It requires four user-defined parameters: interface thickness ($\varepsilon$), artificial time step ($\Delta\tau$), steady-state tolerance, and maximum iterations.
2. Geometric Reinitialization (Novel): A new method proposed in this work, adapted from Level-Set literature (Mut et al., Ausas et al.). It reconstructs the interface geometry using a marching cube method, computes signed distances to the interface for the mesh degrees of freedom, and applies local/global volume corrections. This method relies on a two-parameter framework: interface thickness ($\varepsilon$) and maximum redistanciation distance ($d_{max}$).
3. Projection-based Reinitialization: A simpler method (Aliabadi and Tezduyar) that projects the phase indicator into a sharper space using a piece-wise analytical function. It requires two parameters: an iso-level ($c$) and a sharpening parameter ($\alpha$).

The study evaluates these methods using three application cases:

• Rising Bubble: A 3D extension of the Hysing et al. benchmark, assessing bubble shape, rise velocity, volume conservation, and sphericity.
• Capillary Migration: A droplet migrating due to a surface tension gradient (Marangoni effect), testing the ability to capture tangential interface motion and analytical migration velocity.
• Rayleigh-Plateau Instability: A capillary jet breakup scenario used to assess spatial convergence of surface area and volume.

Key Results

• Performance of PDE-based vs. Geometric Methods: Both the PDE-based and geometric methods achieved high-quality, spatially converged results that agreed well with benchmark and analytic solutions for the rising bubble and capillary migration cases.
  • The PDE-based method showed high sensitivity to parameter selection; optimal values for interface thickness and time steps were found to be case-dependent. It also exhibited artificial displacement of the interface, leading to jumps in sphericity and volume evolution.
  • The Geometric method provided results of comparable quality but with a more robust two-parameter framework that remained consistent across different cases. It maintained the interface position more accurately, resulting in smoother sphericity evolution and better volume conservation, particularly at higher reinitialization frequencies. However, it was noted to be more sensitive to low reinitialization CFL numbers in the capillary migration case.
• Failure of Projection-based Method: The projection-based approach failed to capture complex 3D interfacial dynamics. In the rising bubble case, it resulted in significant volume loss (up to 25%) and strong oscillations in sphericity. In the capillary migration case, it failed to capture the migration velocity entirely and produced strong radius oscillations. Consequently, this method was excluded from the Rayleigh-Plateau instability analysis.
• Spatial Convergence: In the Rayleigh-Plateau case, both the PDE-based and geometric methods demonstrated spatial convergence. The geometric method showed slightly better accuracy for volume and surface area on coarser meshes compared to the PDE-based approach.

Significance and Claims
The paper claims that while state-of-the-art PDE-based reinitialization can yield high-quality results, its effectiveness is heavily dependent on the selection of four hyper-parameters, which vary by case. In contrast, the proposed geometric reinitialization method offers a robust two-parameter framework that delivers consistent, high-quality results across diverse capillary flow scenarios without the need for case-specific tuning.

The authors emphasize that meaningful comparison of reinitialization methods requires sensitive metrics based on interface shape (e.g., sphericity, radius distribution) rather than just global volume, as volume conservation alone may not reveal interface distortion. The study concludes that the geometric method is a promising, robust alternative for complex multi-physics flows involving significant interfacial deformations, such as those found in laser powder bed fusion, where Marangoni effects and evaporation-induced pressure jumps are present.