On buoyancy in disperse two-phase flow and its impact on well-posedness of two-fluid models

This paper resolves the long-standing controversy over buoyancy closures in disperse two-phase flow by deriving a unique, approximation-free closure that correctly attributes all fluid stresses to the background flow (except Reynolds stress), thereby eliminating Hadamard instabilities and ensuring the linear well-posedness of two-fluid models.

The Gist

The Big Picture: The "Ghost" in the Machine

Imagine you are trying to predict how a crowd of people (particles) moves through a flowing river (fluid). Scientists use computer models called "Two-Fluid Models" to do this. They treat the river and the crowd as two separate, intermingling fluids.

For decades, these models have had a major glitch. When scientists tried to simulate them, the math would sometimes explode into nonsense. Small ripples in the data would grow infinitely fast, making the simulation crash. This is called an "ill-posed" problem.

The scientists in this paper discovered that the culprit wasn't the math itself, but a specific ingredient in the recipe: Buoyancy.

The Old Recipe: A Flawed Assumption

To understand the fix, we need to understand the old way of thinking.

The Analogy: The Submarine and the Ocean
Imagine a submarine underwater. It feels a force pushing it up. This is buoyancy. In a calm ocean, this force is simply the weight of the water the submarine displaces.

But what if the ocean is turbulent? What if the water is churning, swirling, and pushing against the submarine from all sides?

• The Old View: Scientists assumed that the "background" water pushing on the submarine was just the average water. They thought, "Let's just take the average pressure and flow of the whole ocean and apply it to the submarine."
• The Problem: This is like trying to describe a hurricane by looking at the average wind speed of the whole planet. You miss the violent, local gusts that actually hit the submarine.

In the old models, scientists included a "fake" stress (called Reynolds stress) in their calculation of this background push. They thought the churning turbulence of the water helped push the particles up or down.

The New Discovery: The "Low-Pass Filter"

The authors of this paper (Zhu, Chen, et al.) realized the old recipe was wrong. They used super-computers to simulate individual particles moving through fluid, acting like a high-definition camera watching every single drop of water.

The Experiment: The "Horizontal Settling"
They created a thought experiment: Imagine a single ball floating in a fast-flowing, turbulent river.

• Old Theory Prediction: The turbulence (swirling water) should push the ball. The model predicted the ball would move with the flow because the "turbulent push" cancels out the drag.
• What Actually Happened: The ball moved against the flow, just like it would in calm water. The turbulence didn't push the ball; it just made the water around it messy.

The Conclusion:
The "background flow" that creates buoyancy is not the average, churning mess. It is the smooth, underlying flow that would exist if the particle weren't there to disturb it.

• The Metaphor: Think of the water flow as a smooth sheet of glass. The particle is a pebble on top. The "buoyancy" is the pressure of the glass sheet. The turbulence is just the pebble vibrating the glass. The vibration doesn't change the pressure of the glass sheet itself.

The "Magic" Fix: The Low-Pass Filter

So, how do they fix the math so it doesn't explode?

They introduced a concept called a "Low-Pass Filter."

The Analogy: The Noise-Canceling Headphones
Imagine you are trying to listen to a song (the physics of the fluid), but there is static noise (tiny, high-frequency mathematical errors) everywhere.

• Old Models: They tried to listen to every frequency, including the high-pitched static. This caused the system to scream (explode).
• The New Model: They put on noise-canceling headphones. They filter out the "high-frequency" noise. They only listen to the smooth, low-frequency parts of the flow.

In physics terms, this filter says: "If the wave of movement is smaller than the particle itself, ignore it."

• If a ripple in the water is smaller than the size of the pebble, the pebble doesn't "feel" it as a buoyant force. It just smooths it out.

Why This Matters

1. It Stops the Math from Exploding: By filtering out the tiny, chaotic ripples that are smaller than the particles, the equations become stable. The "Hadamard instability" (the infinite growth of errors) disappears.
2. It's More Accurate: The old models were wrong about how turbulence affects buoyancy. The new model correctly says that turbulence doesn't push particles up or down; only the smooth, average flow does.
3. It Works for Everything: Whether you are modeling sand in a river, bubbles in a soda, or blood cells in a vein, this new rule makes the computer models stable and reliable.

The Takeaway

For 60 years, scientists tried to fix broken fluid models by adding more complicated "band-aids" (extra friction terms, etc.).

This paper says: "Stop adding band-aids. The recipe itself was wrong."

They found that buoyancy is a "smooth" force, not a "choppy" one. By filtering out the choppy, tiny details that don't physically matter to the particle, they created a model that is mathematically perfect and physically real. It's like realizing that to predict how a boat floats, you don't need to count every single molecule of water; you just need to understand the smooth surface of the lake.

Technical Summary

1. Problem Statement

The paper addresses a long-standing controversy in the modeling of disperse two-phase flows (mixtures of fluid and particles, droplets, or bubbles): How to correctly define the generalized-buoyancy force density in the momentum balance equations of two-fluid models.

• The Core Issue: In two-fluid models, the interfacial force coupling the fluid and dispersed phases is typically decomposed into a "generalized-drag" force and a "generalized-buoyancy" force. While the drag force is well-understood, the physical closure for buoyancy remains debated. Specifically, there is disagreement over which stress components of the background flow contribute to buoyancy.
• The Controversy: Existing closures differ on whether pseudo-stresses (specifically the Reynolds stress $\boldsymbol{\sigma}^f_{Re}$ and the fluid-particle interaction stress $\boldsymbol{\sigma}^f_s$) should be attributed to the background flow stress responsible for buoyancy.
  • Anderson & Jackson (1967)/Jackson (2000): Proposed that all pseudo-stresses contribute to buoyancy.
  • Zhang et al. (2007): Proposed that only the fluid-phase-averaged stress $\langle \boldsymbol{\sigma} \rangle_f$ contributes.
  • Revil-Baudard & Chauchat (2013): Proposed that the interaction stress $\boldsymbol{\sigma}^f_s$ contributes, but the Reynolds stress $\boldsymbol{\sigma}^f_{Re}$ does not.
• The Consequence: The incorrect modeling of buoyancy is widely recognized as the primary cause of Hadamard instabilities (ill-posedness) in two-fluid models, where small-wavelength perturbations grow unboundedly. Most previous attempts to fix this involved adding artificial dissipation terms unrelated to buoyancy, rather than correcting the buoyancy term itself.

2. Methodology

The authors employ a multi-faceted approach combining rigorous mathematical derivation, physical thought experiments, and high-fidelity numerical simulations to derive a unique, consistent buoyancy closure.

• Theoretical Framework:
  • They start from the Maxey-Riley-Gatignol (MRG) equation for a single sphere, which decomposes hydrodynamic forces into background flow contributions (buoyancy) and disturbance flow contributions (drag, virtual mass, history).
  • They utilize exact micromechanical averaging (based on P¨ahtz et al., 2026) to derive macroscopic two-fluid momentum balances without small-particle approximations.
  • They introduce a physical criterion: The buoyancy closure must allow the decomposition of the average fluid momentum flux such that the net flux into the fluid phase and the dispersed phase are driven by the same background stress tensor.
• Numerical Simulations:
  • Particle-Resolving Simulations (PRS): Using the Immersed Boundary Method (IBM), the authors simulate:
    1. Viscous Sediment Transport: A statistically steady flow of particles down a slope to isolate the vertical buoyancy force.
    2. "Horizontal Settling": A single particle in a pressure-driven flow (laminar and turbulent) with no gravity, where buoyancy arises from shear stress gradients. This setup specifically tests the contribution of Reynolds shear stress.
• Thought Experiments:
  • A dilute Stokesian suspension analysis to derive the exact form of the buoyancy force in the limit of vanishing particle Reynolds numbers, testing the consistency of existing closures against the MRG equation.
  • A complete rest scenario in a density-stratified fluid to test the consistency of small-particle approximations.

3. Key Contributions

A. Derivation of the Unique Consistent Closure

The authors derive a unique buoyancy closure that satisfies their physical criterion and is consistent with the MRG equation.

• The Result: The generalized-buoyancy force is driven by the effective mechanical fluid phase stress ($\boldsymbol{\sigma}^f = \beta_f \langle \boldsymbol{\sigma} \rangle_f + \boldsymbol{\sigma}^f_s$).
• Reynolds Stress Exclusion: Crucially, the Reynolds stress ($\boldsymbol{\sigma}^f_{Re}$) does NOT contribute to the buoyancy force. This falsifies the "Jackson closure" and supports the hypothesis of Revil-Baudard & Chauchat (2013) but extends it beyond the small-particle limit.
• Generalization: Unlike previous closures that rely on the "small-particle approximation" (assuming particle size $\ll$ flow gradient scale), the authors derive a closure valid for arbitrarily large particles. This involves a smoothing operator $\mathcal{L}$ that averages stresses over the particle volume.

B. The Low-Pass Filter Property

The derived closure introduces a mathematical operator $\mathcal{L}$ (defined via volume averaging over the particle domain).

• In Fourier space, this operator acts as a low-pass filter with a transfer function $M_K \sim (R|\mathbf{k}|)^{-4}$ for large wavenumbers $|\mathbf{k}|$.
• This means the buoyancy force is attenuated at short wavelengths (scales comparable to or smaller than the particle size).

C. Resolution of Ill-Posedness

The authors demonstrate that the ill-posedness (Hadamard instability) of two-fluid models is not an inherent flaw of the physics but an artifact of using incorrect closures (specifically those lacking the smoothing operator $\mathcal{L}$).

• By employing the new closure, the buoyancy term naturally suppresses high-wavenumber growth.
• Result: Even simplistic two-fluid models (without artificial dissipation) become linearly well-posed when using this closure, provided standard viscous damping and effective dispersed-phase pressure are included.

4. Key Results

1. Validation against Simulations:

   • In the viscous sediment transport simulation, the data confirmed that the buoyancy force aligns with the divergence of the effective stress $\boldsymbol{\sigma}^f$, not just $\langle \boldsymbol{\sigma} \rangle_f$.
   • In the horizontal settling simulation (turbulent flow), the Reynolds shear stress gradient was dominant. The simulations showed that the particle velocity difference matched predictions where Reynolds stress does not contribute to buoyancy. This explicitly falsified the Jackson closure, which predicts that Reynolds stress gradients would cancel the pressure gradient, leading to zero relative velocity.

2. Falsification of Existing Closures:

   • Zhang et al. (2007): Incorrect because it ignores the interaction stress $\boldsymbol{\sigma}^f_s$.
   • Jackson (2000): Incorrect because it includes the Reynolds stress $\boldsymbol{\sigma}^f_{Re}$ in the buoyancy term.
   • Small-particle approximations: Inconsistent for systems at rest in density-stratified fluids and fail to prevent ill-posedness.

3. Virtual Mass Force:

   • The authors also show that the virtual-mass force, when modeled consistently with the background flow averaging (using the operator $\mathcal{L}$), does not introduce ill-posedness, contrary to previous beliefs.

5. Significance

• Theoretical Breakthrough: The paper resolves a decades-old debate by providing a first-principles derivation of the buoyancy closure. It establishes that buoyancy in two-phase flow is fundamentally a stress-gradient phenomenon involving the effective mechanical stress, excluding Reynolds stresses.
• Solving the Ill-Posedness Problem: It provides a first-principle-based solution to the Hadamard instability problem. Instead of adding ad-hoc regularization terms, the correct physical modeling of buoyancy (including particle-size smoothing) naturally stabilizes the equations.
• Practical Implications:
  • For large-scale simulations (grid size $\gg$ particle size), the standard small-particle approximation (Revil-Baudard & Chauchat, 2013) remains sufficient.
  • For high-resolution simulations (grid size $\approx$ particle size) or theoretical stability analyses of short wavelengths, the new closure with the smoothing operator $\mathcal{L}$ is mandatory to ensure physical accuracy and numerical stability.
• Impact on Sediment Transport: The findings suggest that standard models for sediment transport thresholds on slopes may need revision, as the direction of the buoyancy force in turbulent flows is normal to the bed, not aligned with gravity as previously assumed in some models.

In summary, this work redefines the fundamental physics of buoyancy in multiphase flows, proving that the "ill-posedness" of two-fluid models was a consequence of mathematical oversimplification rather than a physical inevitability, and offers a rigorous, consistent closure to fix it.