Lattice Boltzmann framework for multiphase flows by Eulerian-Eulerian Navier-Stokes equations

This paper introduces a novel, dimension-independent Lattice Boltzmann framework that solves Eulerian-Eulerian multiphase flow equations with large density ratios and realistic drag coefficients without finite difference corrections, demonstrating excellent agreement with traditional solvers and promising efficient implementation on high-performance computing systems.

The Gist

Imagine you are trying to predict how a crowd of people (a gas) and a crowd of cars (a liquid) move through a city street together. They bump into each other, push each other, and flow at different speeds. This is what scientists call a multiphase flow.

For decades, simulating this on computers has been like trying to solve a massive, tangled knot of equations. The traditional method (Finite Difference) is like a very careful, slow accountant who checks every single number against a ledger. It's accurate, but it's slow and hard to scale up when you want to simulate a whole city instead of just one street.

This paper introduces a new, faster way to do this using a method called Lattice Boltzmann (LBM). Think of LBM not as an accountant, but as a swarm of tiny, energetic ants. Instead of calculating the whole flow at once, each ant only looks at its immediate neighbors, bumps into them, and moves on. Because every ant works independently, you can have millions of them working at the same time on supercomputers.

Here is the breakdown of what this paper achieved, using simple analogies:

1. The Big Problem: The "Knot" of Two Fluids

Usually, when we simulate gas and liquid together, the math gets messy.

• The Density Gap: Gas is light as a feather; liquid is heavy as a rock. When you try to simulate them together, the computer gets confused because the numbers are so different (like trying to balance a feather and a boulder on a scale).
• The "Correction" Crutch: Most LBM methods needed a "crutch"—a special mathematical patch (Finite Difference correction) to keep the simulation from falling over. But using a crutch slows down the ants and ruins the "swarm" advantage.

The Paper's Solution: They built a new framework that doesn't need the crutch. It's like teaching the ants to balance the feather and the boulder naturally, without extra help.

2. The New Framework: A Six-Piece Orchestra

To solve this, the authors didn't just use one set of ants. They organized six different teams of ants (six coupled schemes) that all run on the same grid, working together like an orchestra:

1. The Gas Team: Simulates the movement of the gas.
2. The Liquid Team: Simulates the movement of the liquid.
3. The Gas Volume Team: Tracks how much space the gas takes up (is it a bubble or a cloud?).
4. The Liquid Volume Team: Tracks how much space the liquid takes up.
5. The Gas "Source" Team: Calculates where the gas is being pushed or pulled.
6. The Liquid "Source" Team: Calculates where the liquid is being pushed or pulled.

The Magic Trick: These six teams talk to each other instantly. They share information about pressure and speed without needing to stop and do complex calculations. This keeps the simulation fast and stable, even when the gas is 800 times lighter than the liquid (a very difficult scenario).

3. The "Artificial Compressibility" Hack

In the real world, liquids are hard to squish (incompressible). But in the computer world, pretending liquids can be squished slightly makes the math much easier to solve.

• The Analogy: Imagine the liquid is a stiff rubber ball. The authors pretend it's a slightly squishy sponge. This allows the "pressure" to travel through the sponge like a wave, helping the ants figure out where to move.
• The Innovation: They figured out how to do this "squishy" trick for both the gas and the liquid at the same time, ensuring they don't get out of sync. They created a special rule so that even though the gas and liquid are different, they both "feel" the same pressure changes.

4. The Drag Force: The "Crowded Dance Floor"

When gas bubbles move through liquid, they drag the liquid with them. The paper uses a realistic model (Clift, Grace, and Weber) to calculate this drag.

• The Challenge: This drag force changes wildly depending on how many bubbles are there. It's like a dance floor: if there are 2 people, they can dance freely. If there are 500, they bump into each other constantly.
• The Fix: The authors added a "damping" mechanism. When the dance floor gets too crowded (the math gets too "stiff"), the simulation slows down the updates just enough to prevent the ants from tripping over each other, ensuring the simulation stays stable.

5. The Results: A Perfect Match

The authors tested their new "six-team ant swarm" against the old "slow accountant" method.

• The Verdict: The results were nearly identical. The new method matched the old, trusted method perfectly.
• The Bonus: Because the new method doesn't need the "crutch" (finite difference corrections), it is perfectly built for High-Performance Computing (HPC). It can be thrown onto massive supercomputers with thousands of processors, solving problems that would take the old method days to finish, in a fraction of the time.

Why Does This Matter?

This isn't just about math; it's about real-world energy.

• Oil & Gas: Better simulations mean we can design better reactors to extract oil or treat gas more efficiently.
• Nuclear Power: It helps simulate how coolant flows in nuclear reactors, making them safer.
• Chemical Industry: It helps optimize bubble columns used to make everything from plastics to medicines.

In Summary:
This paper is like inventing a new, super-efficient way to organize a massive traffic control system. Instead of having one traffic cop (the old method) trying to direct every car and truck, they created a system where every car has a tiny, smart GPS that talks to its neighbors. It's faster, handles huge crowds (density ratios) better, and doesn't need a manual override (finite difference corrections) to keep things moving smoothly.

Technical Summary

1. Problem Statement

Multiphase flow simulations are critical for energy applications (e.g., oil and gas, bubble column reactors, nuclear reactors). Traditional Computational Fluid Dynamics (CFD) methods, such as Finite Volume (FVM) or Finite Difference (FD), struggle with the Eulerian-Eulerian (two-fluid) approach due to:

• Numerical Instabilities: Coupling between phases can lead to instabilities, especially with large density ratios.
• Singularities: Terms involving $\nabla \ln(\alpha_\phi)$ (where $\alpha$ is volume fraction) become singular as $\alpha \to 0$.
• Pressure Coupling: Solving the pressure Poisson equation (Eq. 15 in the paper) is computationally expensive and difficult to parallelize efficiently.
• LBM Limitations: While the Lattice Boltzmann Method (LBM) is highly parallel and suitable for High-Performance Computing (HPC), standard LBM formulations struggle to solve incompressible multiphase equations directly without introducing non-local finite difference corrections, which degrade LBM's parallel efficiency and locality.

The authors aim to develop a pure LBM framework for Eulerian-Eulerian multiphase flows that:

1. Avoids finite difference corrections.
2. Handles very large density ratios.
3. Incorporates realistic drag coefficient models.
4. Maintains the local, parallel nature of LBM.

2. Methodology

The authors propose a novel framework consisting of six coupled LBM schemes running on the same lattice. The core innovation is the use of artificial compressibility for each phase to avoid solving the global pressure Poisson equation.

A. Governing Equations & Artificial Compressibility

Instead of solving the incompressible continuity equation directly, the authors introduce a pseudo-compressible continuity equation for each phase ($\phi \in \{g, l\}$):
$$\frac{\partial \epsilon_\phi}{\partial t} + \nabla \cdot (\alpha_\phi \mathbf{u}_\phi) = S_\phi$$
where $\epsilon_\phi$ is a function of pressure and $S_\phi$ is a source term representing the divergence of the relative velocity between phases. This allows the pressure to evolve dynamically rather than being solved via a global elliptic equation.

B. The Six LBM Schemes

The framework utilizes six distribution functions ($f$) to solve the system:

1. Momentum & Continuity (Gas): $f_g$ solves the momentum equation and the artificial compressible continuity equation for the gas phase.
2. Momentum & Continuity (Liquid): $f_l$ solves the corresponding equations for the liquid phase.
   • Key Innovation: The two phases share the same pressure gradient but use different equations of state (EOS) to handle density ratios. The liquid phase uses a generalized EOS with a tunable parameter $\phi_l$ to ensure the pressure gradient drives both phases consistently.
3. Volume Fractions: $f_{\alpha g}$ and $f_{\alpha l}$ solve the transport equations for the gas and liquid volume fractions ($\alpha_g, \alpha_l$).
   • Constraint Handling: A Spalding-like renormalization is applied to ensure $0 \le \alpha_\phi \le 1$ and $\nabla \alpha_g = -\nabla \alpha_l$.
4. Continuity Sources: $f_{\beta g}$ and $f_{\beta l}$ compute the source terms ($S_g, S_l$) required for the artificial compressibility equations. These are calculated using a Link-Wise Artificial Compressibility Method (LKS) approach, avoiding explicit finite difference derivatives.

C. Asymptotic Analysis

The authors perform a rigorous Chapman-Enskog-like asymptotic analysis using the equivalent moment system. They demonstrate that under diffusive scaling ($\tau/T = (\lambda/L)^2$), the proposed LBM schemes recover the target Eulerian-Eulerian Navier-Stokes equations with second-order accuracy.

• The analysis proves that viscous stress tensors and volume fraction gradients can be computed directly from the distribution functions without additional finite difference operators.

D. Handling Challenges

• Large Density Ratios ($R \gg 1$): To prevent numerical instability when $R$ is large (e.g., $R=833$), the authors introduce:
  1. Dissipation Terms: Adding a dashpot-like term to the continuity equation to suppress spurious acoustic modes.
  2. Source Smoothing: Updating continuity sources and drag coefficients only periodically (every $n_\gamma$ steps) to allow the system to gradually accommodate changes, preventing stiffness-induced oscillations.
• Realistic Drag Models: The framework incorporates the Clift, Grace, and Weber (CGW) model, which introduces a non-linear dependence of the drag coefficient on the volume fraction.

3. Key Contributions

1. First Pure LBM Eulerian-Eulerian Solver: To the authors' knowledge, this is the first LBM framework to solve the full Eulerian-Eulerian two-fluid model without any finite difference corrections.
2. Six-Coupled Scheme Architecture: A modular design using six distribution functions that ensures efficient implementation on large codes with minimal effort.
3. Scalability: The methodology is dimension-agnostic and designed for High-Performance Computing (HPC) and novel parallel hardware, as it relies entirely on local operations.
4. Robustness: Successfully handles very large density ratios (up to $R \approx 833$) and stiff drag models, which are typically problematic for LBM.

4. Numerical Results & Validation

The method was validated against a reference Finite Difference (FD) solver in a 1D vertical tube test case (representing a natural circulation loop). Four test cases were performed:

• TEST #1 & #2 (Preliminary): Density ratios $R=2$ and $R=5$.
  • Result: Excellent agreement between LBM and FD for velocities, volume fractions, and kinematic pressure.
  • Observation: The LBM correctly captures the divergence-free mixture velocity while allowing individual phase velocities to diverge, a physical requirement often missed by simpler models.
• TEST #3 (Advanced): Very large density ratio $R \approx 833$.
  • Result: The stabilization techniques (dissipation and source smoothing) successfully prevented instability. LBM results matched FD results with high fidelity, though minor smoothing was observed in transient profiles.
• TEST #4 (Advanced): Large density ratio ($R \approx 833$) with the stiff CGW drag model.
  • Result: The framework handled the stiffness of the drag model. While minor discrepancies (approx. 8% in volume fraction at the outlet) were noted compared to FD, the overall momentum balance and pressure gradients were consistent. The authors attribute minor errors to the extreme sensitivity of the drag force to small changes in volume fraction at high density ratios.

5. Significance

This work represents a significant step forward in Computational Fluid Dynamics (CFD) for multiphase flows.

• HPC Readiness: By eliminating non-local finite difference corrections, the method fully exploits the locality of LBM, making it ideal for massive parallelization on modern supercomputers.
• Industrial Application: The ability to simulate high-density-ratio flows (gas-liquid) with realistic drag models opens new avenues for optimizing industrial processes like bubble column reactors and gas treatment systems.
• Theoretical Advancement: It bridges the gap between the kinetic theory foundation of LBM and the macroscopic two-fluid models used in traditional CFD, proving that complex multiphase systems can be solved using purely kinetic schemes.

In conclusion, Piredda and Asinari have provided a robust, scalable, and accurate LBM framework that overcomes historical limitations in multiphase flow simulation, paving the way for high-fidelity simulations on next-generation computing architectures.