Adaptive finite volume-particle method for free surface flows

This paper introduces an adaptive finite volume-particle method (AFVPM) that synergistically combines an Eulerian finite volume approach for bulk flow regions with a Lagrangian smoothed particle hydrodynamics formulation for free surface tracking, utilizing a dynamic conversion strategy and buffer region algorithm to achieve superior accuracy and efficiency in simulating complex free surface flows compared to full SPH methods.

The Gist

Imagine you are trying to simulate a massive ocean wave crashing into a ship, or water pouring into a bucket. To do this on a computer, you have to break the water down into tiny pieces and calculate how they move. But here's the problem: water is tricky.

• In the middle of the ocean (the "bulk"), the water moves smoothly and predictably.
• At the surface (the "free surface"), the water is chaotic. It splashes, breaks, forms bubbles, and reconnects.

The paper introduces a new super-smart computer method called AFVPM (Adaptive Finite Volume-Particle Method) that solves this problem by using two different tools for two different jobs, and it switches between them automatically as the water moves.

Here is the breakdown using simple analogies:

1. The Two Tools: The Grid vs. The Marbles

To understand the new method, we first need to understand the two old ways of doing things:

• The Grid Method (FVM): Imagine the water is trapped inside a giant, invisible chessboard. The computer calculates how the water moves from one square to the next.

  • Pros: It's incredibly fast and efficient for calm, deep water.
  • Cons: It's terrible at the edges. If a wave crashes and breaks into a million tiny droplets, the chessboard gets messy. The "squares" can't easily represent a jagged, breaking wave without blurring it out.

• The Marble Method (SPH): Imagine the water isn't in squares, but is made of billions of tiny marbles floating freely. Each marble knows where its neighbors are.

  • Pros: It's amazing at the surface. If a wave breaks, the marbles just fly apart naturally. No blurring.
  • Cons: It's very slow. Because the marbles move around, the computer has to constantly ask, "Who is my neighbor?" for every single marble. Doing this for the whole ocean is computationally expensive (like trying to track every person in a stadium individually).

2. The Innovation: The "Smart Switch"

The authors of this paper realized: "Why use the slow marble method for the whole ocean when we only need it for the messy surface?"

They created a hybrid system that acts like a smart traffic controller:

• Deep Water (The Highway): In the calm, deep parts of the water, the system uses the Grid (Chessboard). It's fast, efficient, and handles the bulk flow perfectly.
• The Surface (The Construction Zone): Wherever the water gets messy—splashing, breaking, or hitting a ship—the system instantly switches to the Marbles. It turns the grid squares into free-floating marbles to handle the chaos.
• The Switching: As the wave moves away and calms down, the marbles turn back into grid squares. As a new wave approaches, the grid turns into marbles.

The Analogy:
Think of it like a video game.

• When you are far away from the action, the game uses a low-resolution, fast-loading map (The Grid) to save your computer's power.
• The moment you get close to an explosion or a complex battle, the game instantly loads the high-resolution, detailed graphics (The Marbles) so you can see the sparks and debris clearly.
• AFVPM does this for water physics, but it happens automatically and seamlessly.

3. The "Buffer Zone": The Diplomatic Handshake

One of the hardest parts of mixing these two methods is the border where the Grid meets the Marbles. If you just put them next to each other, data might get lost, or the simulation might crash (like trying to pour water from a square cup into a round bowl without spilling).

The authors invented a "Buffer Zone" (or a diplomatic handshake):

• They create a special transition area where the Grid and Marbles talk to each other.
• They use "ghost marbles" and "ghost squares" to make sure the data flows smoothly across the border.
• This ensures that when a wave moves from the calm deep water into the splashing surface, the physics remains perfect, with no glitches.

4. Why Does This Matter?

The paper tested this method on several scenarios:

• Still Water: It worked perfectly.
• Dam Breaking: A huge wall of water collapsing. The method captured the splash perfectly while staying fast.
• Ship Cruising: Simulating a ship cutting through water.
• Dropping a Cylinder: Watching a heavy object hit the water and create shockwaves.

The Result:
The new method was about 1.5 times faster than using the slow "Marble" method for everything, but it was just as accurate (or even better) at capturing the messy details of the splashing water.

Summary

The paper presents a smart, adaptive water simulator. Instead of using one expensive tool for the whole job, it uses a fast, efficient tool for the calm parts and a precise, detailed tool for the messy parts, switching between them instantly. This allows scientists and engineers to simulate complex water events (like tsunamis, ship crashes, or filling a pool) much faster and more accurately than before.

Technical Summary

1. Problem Statement

Free-surface flows involving complex interface dynamics (breaking, fragmentation, reconnection) are challenging to simulate efficiently and accurately.

• Mesh-based methods (e.g., FVM, VOF, Level Set): Offer high computational efficiency and accuracy in bulk flow regions but struggle with interface reconstruction on fixed grids, often introducing numerical diffusion that smears the interface.
• Particle-based methods (e.g., SPH): Naturally handle large interface deformations without explicit tracking but suffer from high computational costs due to neighbor searching at every time step and consistency issues.
• Existing Hybrid Methods: Previous attempts to combine these approaches often lack dynamic adaptability, use irreversible conversions (e.g., elements to particles), or rely on static transition layers that restrict mesh-to-particle ratios, limiting overall efficiency and accuracy.

The authors propose a solution to dynamically combine the efficiency of Eulerian meshes with the robustness of Lagrangian particles, adapting the discretization method based on the local flow physics.

2. Methodology: Adaptive Finite Volume-Particle Method (AFVPM)

The AFVPM framework synergistically integrates an unstructured-mesh Finite Volume Method (FVM) with a Weakly Compressible Smoothed Particle Hydrodynamics (SPH) approach.

A. Governing Equations & Solvers

1. Lagrangian Region (SPH):
   • Used for the free surface and interface regions.
   • Solves the Navier-Stokes equations in Lagrangian form with a weakly compressible equation of state ($p = C_0^2(\rho - \rho_0)$).
   • Incorporates a numerical diffusive term to stabilize pressure and a Particle Shifting Technique (PST) to regularize particle distribution.
   • Uses a C-2 Wendland kernel function.
2. Eulerian Region (FVM):
   • Used for the bulk flow regions.
   • Employs an Isothermal Gas-Kinetic Scheme (GKS) based on the BGK equation.
   • GKS calculates fluxes by combining inviscid and viscous components from the gas distribution function at cell interfaces, allowing for high-order accuracy and natural handling of gravity effects.
   • Modified to accommodate non-ideal equations of state (specifically the weakly compressible EOS used in SPH) by adding a pressure correction term to the numerical flux.

B. Adaptive Conversion Strategy

The computational domain is divided into Cartesian mesh blocks (the unit of conversion) rather than individual cells. Blocks are categorized as:

• Interior: Fluid-filled, far from the free surface (Solved by GKS).
• Interface-Mesh: Fluid-filled, adjacent to particle regions (Solved by GKS).
• Interface-Air (Buffer): Contains real particles near the interface; acts as a transition zone.
• Void: Empty regions.

Two-Way Dynamic Conversion Mechanisms:

1. Particle to Mesh: If a deactivated block is far from the free surface ($> DX$), it activates. Real particles are deleted, and field variables are interpolated from neighbors to initialize the mesh cells.
2. Mesh to Particle: If an interface-mesh block approaches the free surface ($< DX$), it deactivates. Buffer particles are converted to real particles.
3. Buffer Management:
   • Buffer Particles: Generated in interface-mesh blocks to support the kernel function of real particles near the boundary. They are updated via interpolation from real particles and active mesh cells.
   • Buffer Cells: Mesh cells in interface-air blocks where field variables are interpolated from neighboring real/buffer particles to calculate fluxes for the GKS solver.

C. Data Communication

A robust interface algorithm ensures seamless data exchange:

• Particle $\to$ Mesh: Field variables in buffer cells are reconstructed using weighted sums of neighboring real and buffer particles.
• Mesh $\to$ Particle: Buffer particles are updated using interpolated values from neighboring real particles and active mesh cells.
• Conservation: The conversion strategy maintains mass and momentum conservation across the interface.

3. Key Contributions

1. Novel Adaptive Framework: First implementation of a fully adaptive, two-way conversion strategy between Eulerian FVM and Lagrangian SPH, allowing the solver to switch discretization types dynamically based on the free surface location.
2. Buffer Region Algorithm: Introduction of a specific "buffer region" concept (both buffer particles and buffer cells) to ensure seamless data communication and kernel support across the mesh-particle interface, preventing numerical instability.
3. Isothermal GKS with Gravity: Adaptation of the Gas-Kinetic Scheme for isothermal flows with gravitational effects and arbitrary equations of state, enabling high-fidelity flux calculation in the mesh regions.
4. Block-Based Conversion: Utilization of mesh blocks (rather than individual cells) as the conversion unit to ensure sufficient resolution (at least two layers of cells) and stability during the transition.

4. Numerical Results & Validation

The method was validated against five benchmark cases:

1. Still Tank: Verified hydrostatic accuracy; results matched analytical solutions perfectly.
2. Dam Breaking: Demonstrated dynamic conversion of mesh/particle regions. Pressure history at the wall matched experimental data and reference SPH solutions.
3. Ship Cruising: Successfully simulated a moving body on a free surface, capturing bow spray, reattachment, and stern waves. The mesh/particle partition adapted smoothly to the ship's motion.
4. Body Entry (Cylinder): Compared against full SPH. AFVPM showed superior free-surface resolution and symmetry, with better handling of pressure wave propagation.
5. Water Filling: A complex case involving injection, wall impact, fragmentation, and reconnection. AFVPM captured fine flow details (e.g., rolled water impacting) better than full SPH.

Computational Efficiency:

• Speedup: AFVPM achieved approximately 1.4x to 1.6x speedup (up to 150% efficiency gain) compared to full SPH across all test cases.
• Comparison: In a still tank case, full SPH took nearly 3 times longer than a full GKS method on the same grid, highlighting the efficiency of the mesh-based approach for bulk flows.

5. Significance

This study presents a significant advancement in interfacial flow modeling by effectively balancing accuracy and computational efficiency.

• Robustness: It overcomes the limitations of pure SPH (high cost, consistency issues) and pure FVM (interface smearing, reconstruction complexity).
• Scalability: The block-based adaptive strategy makes the method suitable for large-scale simulations where the free surface moves significantly.
• Future Impact: The framework provides a solid foundation for next-generation hybrid solvers, with potential extensions to multiphase flows and parallel computing for industrial applications in ocean engineering and fluid-structure interaction.