Simulating the interaction between fluids and solid objects is a cornerstone of computer graphics and computational engineering. However, a notoriously difficult challenge arises when the solid objects are thin, "codimensional" structures—such as cloth, shells, or thin rods.
As noted by Li et al., these objects are mathematically represented as 2D surfaces or 1D curves embedded in a 3D world. Because they lack physical volume, traditional physics engines frequently suffer from "volumeless collision failure," allowing fluids to erroneously leak through solid boundaries. Recent research has converged on a shared goal: building robust, leak-proof boundary conditions that guarantee non-penetration even under extreme, high-velocity conditions.
The Volumeless Leakage Problem
To understand why thin structures leak, we have to look at how fluids are traditionally simulated. Fluid solvers typically discretize space into a fixed grid (an Eulerian approach) or track the fluid as a collection of moving points (a Lagrangian approach).
In a grid-based solver, a thin shell can easily fall entirely between grid cells. If the solver only checks for velocity alignment at discrete grid points, the thin solid becomes effectively "invisible" to the fluid, as highlighted by Sun et al..
In a particle-based solver, high-velocity fluid particles can easily "tunnel" straight through a zero-thickness boundary within a single time step. When boundaries leak, the simulation loses mass, pressure drops artificially, and the physical coupling destabilizes, leading to unrealistic tangles or explosions.
Traditional Grid/Particle Approach Position-Level / Barrier Approach
Fluid Particle Fluid Particle
│ │
▼ (High velocity) ▼ (Repelled by barrier)
─────────────────── (Thin Shell) ▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲ (Barrier Zone)
│ ─────────────────── (Thin Shell)
▼ (Leaked!)
To solve this, modern techniques move away from simple velocity-coherence checks and instead enforce non-penetration through geometric, positional, and kinetic guarantees.
Position-Level and Barrier Constraints
One powerful way to prevent leakage in particle and Lagrangian systems is to enforce non-penetration directly at the positional level, rather than relying on velocity corrections after a collision has already occurred.
Sun et al. developed a unified optimization framework that simultaneously solves for fluid incompressibility, solid elasticity, and geometric non-penetration. By enforcing non-penetration as a hard positional constraint, their method ensures that fluids cannot cross the boundaries of shells and rods, even during high-speed impacts or intricate deformations.
Another approach introduces mathematical "barriers" that act before a particle can cross a boundary. Li et al. introduced the Contact Proxy Splitting method, which utilizes barrier-function constraints. These functions act like ultra-stiff, short-range forces that rapidly repel fluid particles as they approach a thin solid. By splitting the contact math into an efficient proxy phase, they achieve a leak-proof guarantee without the prohibitive computational costs usually associated with strict optimization.
For Smoothed Particle Hydrodynamics (SPH), a simpler but highly effective geometric strategy is to change how the boundary itself is represented. Kulp et al. demonstrated that representing thin, rapidly moving walls (such as cardiac valves) as thickened layers of boundary particles ensures robust containment. This artificial thickness prevents fluid particles from finding gaps to tunnel through during rapid geometric changes.
Double-Sided Bounce-Back in Kinetic Solvers
For grid-based kinetic frameworks like the Lattice Boltzmann Method (LBM), the challenge manifests in "cut-cells"—cells that are partially intersected by a solid boundary geometry.
Traditional LBM struggles with thin structures because fluid populations can slip past the sub-grid boundary representation. To resolve this, Lyu et al. proposed a double-sided bounce-back scheme. In standard LBM, fluid distributions "bounce back" when they hit a wall. By extending this rule to all cells intersected by a solid boundary—regardless of whether they are technically classified as fluid or solid cells—the scheme creates a continuous, impenetrable barrier on both sides of the thin shell, completely blocking fluid leakage.
Standard Cut-Cell LBM Double-Sided Bounce-Back
Fluid Solid Fluid Solid
┌───────┐ ┌───────┐ ┌───────┐ ┌───────┐
│ ─► │ ╲ │ │ │ ◄─► │ ╲ │ ◄─► │ ◄── Bounce-back
│ │ ╲│ │ │ │ ╲│ │ applied to
└───────┘ └───────┘ └───────┘ └───────┘ ALL cut-cells
▲ ▲
Thin Boundary Thin Boundary
(Fluid leaks through gaps) (Impenetrable barrier)
Building on kinetic foundations, Li and Desbrun paired a hybrid moving bounce-back scheme with a smoothed pressure filter. This combination suppresses the artificial pressure oscillations that typically emerge near moving boundaries, stabilizing the simulation.
Furthermore, handling the exact forces acting on these boundaries is critical. Hsieh and Tsai showed that fixing the computational frame of reference directly to the interacting body allows for highly precise surface stress calculations. This precision prevents numerical instability even when the solid object is extremely light relative to the surrounding fluid.
Enabling Accurate Fluid Reconstruction
Developing leak-proof physics solvers is not just important for forward simulations; it is also a vital piece of the puzzle for inverse problems, such as reconstructing 3D fluids from video.
When algorithms attempt to infer fluid motion from visual data, they must model how the fluid interacts with obstacles. As Franz et al. point out, older global transport formulations frequently break down because they lack support for complex 3D boundaries or immersed objects, leading to physically impossible reconstructions where smoke or water passes through containers.
Modern differentiable pipelines leverage robust boundary physics to bridge the gap between computer vision and deep learning. For instance, the FluidNexus framework introduced by Gao et al. couples generative video diffusion models with differentiable physics simulations to predict fluid behavior from a single camera view. For these models to successfully simulate interactive scenarios—like adding a new thin obstacle into a pre-recorded video—the underlying physics engine must precisely align the fluid's velocity field with the solid's surface.
By guaranteeing that boundaries remain completely leak-proof, these modern coupling frameworks provide the mathematical stability required to make advanced neural-physical pipelines work in the real world.
Key Foundational Work
Fast and versatile fluid-solid coupling for turbulent flow simulation (2021) — contributed 6 concept elements to this direction.
Practical patient-specific cardiac blood flow simulations using SPH (2013) — contributed 2 concept elements to this direction.
A target-fixed immersed-boundary formulation for rigid bodies interacting with fluid flow (2021) — contributed 1 concept elements to this direction.