Adaptive near-contact repulsion in conservative Allen-Cahn phase-field lattice Boltzmann multiphase model

This paper introduces a fully local, adaptive repulsive flux within a conservative Allen-Cahn phase-field lattice Boltzmann model to effectively prevent spurious coalescence in multiphase flow simulations by dynamically adjusting interaction strength based on estimated local film thickness, thereby ensuring robust and physically consistent near-contact dynamics without sacrificing computational efficiency.

The Gist

The Big Problem: The "Ghostly" Merge

Imagine you are watching a video game where two soap bubbles float toward each other. In the real world, if they get very close, a tiny, invisible layer of air (or liquid) gets trapped between them. This layer acts like a cushion, slowing them down and often making them bounce off each other instead of sticking together.

However, in computer simulations of fluids, this tiny layer is often too thin to see. Because the computer can't "see" the air cushion, it thinks the two bubbles have touched. As a result, the simulation makes them instantly merge into one big blob. This is called spurious coalescence (or "fake merging"). It ruins the simulation because, in reality, those bubbles should have bounced apart.

The Solution: An "Adaptive Airbag"

The authors of this paper have invented a new digital rule to fix this. They call it an Adaptive Near-Contact Repulsion.

Think of it like this:

• The Old Way: Imagine the bubbles have a fixed "force field" around them. If they get too close, the force pushes them apart. The problem is, this force is either too weak (they still merge) or too strong (they bounce off like rubber balls even when they shouldn't). It's a "one-size-fits-all" approach.
• The New Way: The authors gave the bubbles a smart, self-adjusting airbag.
  1. It only wakes up when needed: The airbag stays deflated until the bubbles are very close and facing each other directly (like two people about to bump heads).
  2. It measures the gap: Instead of using a ruler or a complex camera to measure the distance, the simulation looks at the "fuzziness" of the bubble's edge. If the fuzziness of two bubbles overlaps in a specific way, the computer knows, "Ah, there is a thin film here!"
  3. It adjusts its strength: The airbag inflates just enough to push the bubbles apart, but not so much that it blasts them away. If the bubbles are trying to squeeze together hard, the airbag gets stronger. If they are just drifting near each other, the airbag stays soft.

How It Works (The "Magic" Trick)

The paper describes a clever mathematical trick to avoid doing expensive, slow calculations.

Usually, to know if two bubbles are about to merge, a computer has to draw a line from one bubble to the other and count how many pixels are in between. This is like sending a scout to measure the distance—it takes time and slows down the simulation, especially on supercomputers.

The authors' method is like reading a mood ring.

• The simulation looks at the "phase field" (a number that tells the computer if a spot is liquid or gas).
• When two bubbles get close, their "mood rings" overlap.
• The computer uses a simple formula to say, "Based on how much these rings overlap, the gap must be this thin."
• Because this calculation happens instantly at every single point without looking at neighbors, it is incredibly fast and perfect for running on powerful graphics cards (GPUs).

The Results: Bouncing and Swarms

The team tested this new "smart airbag" in two ways:

1. The Head-On Crash: They simulated two droplets crashing into each other. In the real world, they flatten out, stop, and bounce back. Without the new rule, the computer made them merge. With the new rule, the simulation perfectly matched real-life experiments, showing the droplets bouncing apart just like they should.
2. The Bubble Swarm: They simulated 100 bubbles rising in a tank of water. In a normal simulation, these bubbles would constantly merge into giant, unrealistic blobs. With the new rule, the bubbles danced around each other, bumped, bounced, and rearranged themselves without ever sticking together. This created a chaotic, turbulent flow that looked very realistic.

Why This Matters

This research is a big deal for scientists and engineers because:

• It's Fast: It doesn't slow down the computer, allowing for massive simulations of complex systems (like oil flowing through rock or clouds forming in the sky).
• It's Smart: It doesn't need human tweaking. The "airbag" automatically knows how hard to push based on the situation.
• It's Realistic: It finally allows computers to simulate the tricky moment when two fluids get close but don't touch, which is crucial for understanding everything from inkjet printers to climate models.

In short: The authors gave computer simulations a "sixth sense" to feel the invisible cushion between fluids, preventing them from making the mistake of merging when they should bounce.

Technical Summary

1. Problem Statement

In diffuse-interface simulations of multiphase flows (e.g., using phase-field methods), a persistent challenge is the spurious coalescence of interfaces. This occurs when two deformable interfaces (such as droplets or bubbles) approach each other across a thin intervening film that is thinner than the numerical grid resolution.

• Physical Reality: In real systems, thin films drain until they reach a critical thickness where rupture occurs, or they remain stable due to short-range repulsive forces (steric, electrostatic, or disjoining pressures), leading to rebound.
• Numerical Failure: Standard phase-field models cannot resolve nanometric film drainage. Consequently, interfaces often merge prematurely due to numerical diffusion or insufficient resolution, even in regimes where physical systems would bounce.
• Limitations of Existing Solutions: Previous methods to prevent this often rely on non-local geometric procedures (e.g., ray tracing along interface normals) to detect neighboring interfaces. These approaches are computationally expensive, difficult to parallelize, and break the locality required for efficient high-performance computing (HPC).

2. Methodology

The authors propose a fully local, adaptive near-contact interaction (NCI) model integrated into a conservative Allen–Cahn phase-field formulation coupled with a lattice Boltzmann (LB) hydrodynamic solver.

A. Hydrodynamic Solver

• Lattice Boltzmann Method (LBM): Uses a D3Q27 stencil with a thread-safe, recursive regularization scheme. This ensures stability at low viscosities and eliminates race conditions on shared-memory architectures (GPUs).
• Conservative Allen–Cahn (CAC): The interface is tracked via a phase field $\phi \in [0,1]$. The evolution equation includes a compressive flux term to maintain a sharp interface profile without excessive numerical diffusion. The scheme uses a MUSCL-TVD (minmod limiter) for advection to ensure second-order accuracy and stability.

B. The Near-Contact Interaction (NCI) Model

The core innovation is a short-range repulsive body force, $\mathbf{F}_{rep}$, added to the momentum equation. Its key features are:

1. Activation Criteria: The force activates only when two diffuse interfaces are in close proximity and exhibit opposite orientations (normals pointing toward each other). This distinguishes a thin-film configuration from a single curved interface.
2. Analytical Film Thickness Estimation: Instead of ray tracing, the local film thickness $h$ is estimated analytically from the local phase-field value $\phi$ (specifically the interfacial indicator $q = \phi(1-\phi)$) using the equilibrium hyperbolic-tangent profile.
   • Formula: $h(q) \approx \varepsilon \, \text{arcosh}\left(\frac{1}{2\sqrt{q}}\right)$.
3. Self-Tuning Mechanism: The magnitude of the repulsive flux is not a fixed penalty. It is modulated by a weight function $w_h(h)$ that increases as the film thins. The force scales dynamically with the local degree of interface overlap and the hydrodynamic tendency toward coalescence.
4. Symmetry and Locality: The force is constructed to be purely normal and antisymmetric (action-reaction) between partner cells. The entire algorithm is strictly local (bounded neighborhood search), making it ideal for massively parallel GPU implementations.

3. Key Contributions

• Elimination of Non-Local Operations: The method replaces computationally heavy ray-tracing or geometric reconstruction with an analytical mapping based on the phase-field profile, preserving the compact stencil of the LB method.
• Adaptive Barrier: The model acts as a "self-tuning" barrier. Once the repulsive amplitude is sufficient to prevent coalescence, the macroscopic dynamics become insensitive to further increases in the amplitude parameter. The system adjusts the spatial support of the repulsion rather than the force magnitude, ensuring physical consistency.
• Computational Efficiency: The formulation is fully local and thread-safe, enabling efficient execution on modern HPC architectures (GPUs) for large-scale 3D simulations.
• Physical Fidelity: The repulsive force mimics the effect of unresolved disjoining pressures, allowing the simulation to capture the correct transition between coalescence and rebound without resolving the nanometric film.

4. Results

The model was validated through three distinct benchmarks:

A. Droplet Collision Benchmarks

• Setup: Head-on and off-axis collisions of dodecane and water droplets, matching experimental conditions (Huang & Pan, 2021) with varying Weber numbers and impact parameters.
• Outcome: The model successfully reproduced the non-coalescing rebound observed in experiments. It captured the correct deformation levels, rebound timing, and post-impact trajectories.
• Hydrodynamics: Analysis of the velocity fields revealed the correct physical signatures of thin-film sustenance:
  • Suppression and inversion of the film-normal flow (drainage stops and reverses).
  • Formation of stagnation-like zones and recirculation patterns at the film rim.
  • These features confirm the model reproduces the hydrodynamic resistance required for rebound.

B. Self-Tuning Analysis

• The authors demonstrated that once the repulsive amplitude ($A_{rep}$) exceeds a critical threshold, the collision outcome (rebound vs. coalescence) and macroscopic flow features become insensitive to the specific value of $A_{rep}$.
• Force balance analysis showed that the repulsive force in the film remains commensurate with the capillary force driving the interfaces together. The model adapts by changing the volume of the activated region rather than amplifying the force indefinitely.

C. 3D Bubble Swarm (Many-Body Interactions)

• Setup: A swarm of 100 rising bubbles in a periodic 3D domain ($512^3$ lattice units) with high density ratios ($\rho_l/\rho_g = 1000$).
• Outcome: The model successfully prevented spurious coalescence over long integration times, maintaining a dispersed bubble population.
• Pseudo-Turbulence: The simulation captured the emergence of "pseudo-turbulence" (chaotic flow driven by bubble wakes). The energy spectrum of the carrier phase exhibited a $k^{-3}$ power-law decay, a hallmark of bubble-driven turbulence, validating the model's ability to handle complex, many-body interactions.

5. Significance

This work provides a robust, scalable, and physically consistent framework for simulating unresolved near-contact dynamics in multiphase flows. By removing the need for non-local geometric operations, it bridges the gap between mesoscopic phase-field models and the physical reality of thin-film stabilization.

• Applications: The method is particularly valuable for simulating complex industrial and natural phenomena involving dense dispersions, such as emulsification, boiling, cloud physics, and carbon sequestration, where many-body collisions and rebounds are frequent.
• Future Impact: The adaptive nature of the model reduces the sensitivity to parameter tuning, making it a reliable tool for large-scale, high-fidelity simulations on modern supercomputers. Future work will focus on calibrating the repulsive closure against thin-film drainage theory and extending the model to wetting and wall-mediated interactions.