Kinetic closure of turbulence: collision-side modeling beyond the filtered BGK--Boltzmann equation

This paper extends a kinetic closure of turbulence by developing a theoretical framework that addresses the non-Markovian collision dynamics and unresolved subgrid equilibrium in filtered Boltzmann equations through a Chapman--Enskog analysis, ultimately validating the resulting BGK-like closure against lattice Boltzmann simulations and traditional models.

The Gist

Imagine you are trying to predict how a crowd of people moves through a busy train station.

The Old Way (Macroscopic View):
Most scientists look at the crowd from a high balcony. They see the "average" flow of people. But because they can't see every single individual, they have to guess what the hidden, fast-moving people are doing. They usually assume these hidden people just act like a thick, sticky fluid (like honey) that slows things down. This is the standard way of modeling turbulence (chaotic flow) in engineering.

The New Way (Kinetic View):
This paper proposes a different perspective. Instead of looking at the crowd from the balcony, imagine you are standing on the floor with a camera that records every single person's position and speed. This is the "Boltzmann equation" approach.

The authors argue that when you filter this detailed camera footage to create a "coarse" view (ignoring the tiniest, fastest movements), you don't lose the information about how people bump into each other. The information is still there, hidden in the details of the crowd's movement.

Here is the core idea broken down with simple analogies:

1. The "Traffic Jam" Analogy

Think of a highway.

• The Macroscopic View (Old Way): You see the average speed of cars. When traffic gets chaotic, you assume the "missing" cars are just creating extra friction (like a thick fog) that slows everyone down. You model this friction as a new, artificial force.
• The Kinetic View (This Paper): You see that the "missing" cars are actually still driving on the road, just moving in ways you aren't tracking individually. The problem isn't that the cars are missing; it's that your model of how cars collide (interact) is too simple.

2. The "Memory" Problem

The paper says the biggest mistake in current models is assuming that when two particles (or people) collide, they forget everything that happened to them a split second ago. This is called a "Markovian" process (no memory).

The authors show that when you blur the picture (filter the data) to ignore tiny details, the collisions do have a memory. The "blur" creates a lag. The particles remember they just bumped into someone because the averaging process smoothed out the exact moment of impact.

• Analogy: Imagine taking a photo of a fast-moving baseball bat hitting a ball. If you use a slow shutter speed (filtering), the photo shows a blur. If you try to predict the next hit based on that blurry photo, you can't just say "they hit and forgot." The blur itself contains a "ghost" of the impact that needs to be accounted for.

3. The "Dual Problem"

The authors realized that fixing this requires solving two problems at once:

1. The Equilibrium Gap: You need to figure out what the "perfectly calm" state of the crowd looks like after you've blurred the picture, which is different from the calm state of the unblurred picture.
2. The Collision Memory: You need to add a new rule to your model that accounts for the "ghost" of the collisions (the covariance) that the blurring created.

4. The Solution: "Recorrelated" Models

The paper introduces a new mathematical framework called the "Filtered Recorrelated BGK–Boltzmann Equation."

• BGK is a simplified way of calculating collisions (like a rulebook for how people bump into each other).
• Recorrelated means they added a special "memory term" to the rulebook.

Think of it like upgrading a video game physics engine. The old engine assumed that if you smoothed out the graphics, the physics would just get "stickier." The new engine realizes that smoothing the graphics actually changes how the objects bounce, so it adds a specific "re-correction" step to the collision math to fix the bounce.

5. How They Tested It

They didn't just write equations; they built a computer simulation (using a method called Lattice Boltzmann) to test their new rulebook. They ran three famous tests:

• The Taylor-Green Vortex: A swirling, chaotic fluid that breaks down into smaller and smaller swirls.
• The Lid-Driven Cavity: A box where the top lid slides, dragging the fluid inside.
• Flow Past a Cylinder: Wind blowing around a pole.

The Results:
Their new model (called KC-RB, KC-MP, and KC-RR) was better at keeping the "small swirls" (turbulence) alive without making the simulation crash or become too blurry. Compared to the old "Smagorinsky" models (the standard "sticky fluid" approach), their new models kept the chaotic details sharper and more accurate, especially when the computer grid wasn't super high-resolution.

Summary

In short, this paper says: "Don't just guess that turbulence acts like thick honey. Instead, realize that when you ignore the tiny details, the way things collide changes. We found a way to mathematically fix the collision rules so they remember the 'ghost' of the tiny details you ignored, leading to much more accurate simulations of chaotic flows."

Technical Summary

Technical Summary: Kinetic Closure of Turbulence: Collision-Side Modeling Beyond the Filtered BGK–Boltzmann Equation

Problem Statement
The paper addresses the closure problem in Large-Eddy Simulation (LES) and Reynolds-Averaged Navier–Stokes (RANS) when applied within a kinetic framework, specifically the Boltzmann equation (BE). While macroscopic filtered Navier–Stokes equations (FNSE) require modeling unresolved advective fluxes (Reynolds or subgrid-scale stresses), the authors argue that filtering the Boltzmann equation fundamentally alters the closure structure. In the filtered BE, the linear streaming operator commutes with filtering, meaning subgrid-scale (SGS) advective transport is not lost but remains embedded within the filtered distribution function. Consequently, the unresolved physics is concentrated entirely on the collision side.

The authors identify a critical limitation in standard kinetic turbulence approaches: they often conflate the breakdown of the molecular chaos hypothesis (two-particle correlations) with the actual source of closure error in filtered kinetic equations. They argue that for dilute gases in LES/RANS regimes, the primary issue is not the failure of molecular chaos, but the retention of a Markovian collision process at a scale where finite-time filtering induces temporal correlations in the collision product. Standard BGK-type models fail because they assume instantaneous relaxation to an equilibrium constructed solely from filtered moments, ignoring the "collision-covariance source term" generated by the covariance of the collision product under the filter.

Methodology
The paper develops a theoretical framework for a "filtered recorrelated BGK–Boltzmann equation" (FRBGK–BE). The methodology proceeds through the following steps:

1. Theoretical Formulation: The authors derive the filtered kinetic equation by applying a kinetic time filter (width $\Delta t$) followed by a homogeneous LES-type spatial filter (width $\Delta x$). They demonstrate that the time filter induces a velocity-dependent spatial extent ($\Delta x \sim c_t \Delta t$, where $c_t$ is the thermal speed), necessitating a consistent space-time coarse-graining scale.
2. Dual Closure Problem: The analysis reveals a dual closure problem:
   • Distinguishing the filtered fine-grained equilibrium (which contains SGS information) from the equilibrium constructed solely from filtered moments.
   • Modeling the non-Markovian collision dynamics generated by the collision-product covariance.
3. Chapman–Enskog (CE) Analysis: A CE expansion is performed using the ratio of kinetic to macroscopic timescales ($\epsilon = Ma^2/Re_\ell$) as the expansion parameter, rather than an artificial turbulent Knudsen number. This analysis shows that the hydrodynamic limit recovers the FNSE only if the collision-covariance source term satisfies specific constraints related to the advective evolution of the SGS stress tensor.
4. Operational Closures: The authors propose a closure strategy where the unresolved collision action is represented as a relaxation of the "SGS carrier" ($f_{sgs}$), defined as the difference between the filtered fine-grained equilibrium and the resolved equilibrium. Three specific operational realizations are formulated:
   • KC-RB (Residual-Based): Estimates the SGS carrier as the total residual minus the first-order CE reconstruction.
   • KC-MP (Moment-Projected): Projects the SGS correction onto the second-order moment manifold.
   • KC-RR (Recursive Kinetic): Generalizes the recursive regularization philosophy to partition moments into resolved (laminar) and unresolved (fluctuating) components using a recursive manifold.
5. Turbulent Relaxation: A constitutive turbulent relaxation frequency ($\omega_t$) is introduced, modeled phenomenologically using an eddy-viscosity argument based on the SGS kinetic energy, bounded by the molecular relaxation rate.

Key Contributions

• Structural Inversion of Kinetic Closure: The paper establishes that in kinetic LES, the SGS advective information is exactly transported by the linear streaming of the filtered distribution. Therefore, the closure problem is not the reconstruction of missing advective fluxes (as in FNSE) but the modeling of the collision-covariance source term.
• Non-Markovian Collision Modeling: The work explicitly identifies the failure of standard BGK models in LES as the neglect of the collision-product covariance induced by finite-time filtering, rather than the breakdown of molecular chaos.
• Consistent Scaling: The derivation avoids artificial scale separations by using the natural timescale ratio from the nondimensionalized kinetic equation, retaining independent Mach-number dependence in higher-order moments.
• New Kinetic Closures: The paper introduces three concrete kinetic closures (KC-RB, KC-MP, KC-RR) that act on the collision side to dissipate the SGS stress content carried by the distribution function.

Results
The proposed closures were validated using lattice Boltzmann simulations (LBM) on three benchmark cases: the 3D Taylor–Green vortex (TGV), lid-driven cavity flow, and flow past a circular cylinder.

• Taylor–Green Vortex: At coarse resolutions ($N=32, 64$), the kinetic closures (KC-RB, KC-MP) demonstrated smoother energy decay and better-timed enstrophy peaks compared to classical Smagorinsky models (both naive and finite-difference variants). The kinetic closures retained sharper transitional vortical structures. At higher resolutions ($N=128$), the kinetic closures converged toward the standard BGK limit due to the clipping of the SGS relaxation rate.
• Lid-Driven Cavity: The kinetic closures showed improved accuracy in predicting root-mean-square velocity fluctuations and Reynolds shear stresses compared to Smagorinsky and standard recursive regularization (RR) models. KC-RR achieved the lowest aggregate RMS error among the tested models.
• Flow Past Circular Cylinder: The kinetic closures (particularly KC-MP and KC-RR) provided more accurate predictions of the recirculation length and near-wake velocity profiles compared to Smagorinsky and HRR models. They retained a denser field of small-scale vortical structures in the wake, consistent with high-resolution reference data, whereas Smagorinsky and HRR tended to over-damp these structures.

Significance and Claims
The paper claims that its framework resolves a long-standing ambiguity in kinetic turbulence modeling by correctly identifying the location of the closure problem. It argues that previous attempts to model kinetic turbulence often incorrectly treated the problem as a kinetic version of the Boussinesq/Smagorinsky ansatz for advective stresses. Instead, this work demonstrates that the kinetic closure is a collision-side problem involving non-Markovian relaxation.

The authors state that their framework provides a formal route to address this dual collision-side closure problem without introducing additional macroscopic transport equations, as the SGS information is already embedded in the filtered kinetic distribution. The numerical results suggest that these collision-side closures offer improved stability and accuracy over traditional macroscopic-based models (like Smagorinsky) and standard regularization techniques in under-resolved turbulent simulations. The work is limited to isothermal flows, with the extension to energy equations noted as a natural future step.