Add-ons for Lattice Boltzmann Methods: Regularization, Filtering and Limiters

R. A. Brownlee, Jeremy Levesley, David Packwood, Alexander N. GorbanBENTHAM 2013 · DOI · arXiv · OpenAlex

Abstract

We describe how regularization of lattice Boltzmann methods can be achieved by modifying dissipation. Classes of techniques used to try to improve regularization of LBMs include flux limiters, enforcing the exact correct production of entropy and manipulating non-hydrodynamic modes of the system in relaxation. Each of these techniques corresponds to an additional modification of dissipation compared with the standard LBGK model. Using some standard 1D and 2D benchmarks including the shock tube and lid driven cavity, we explore the effectiveness of these classes of methods.

Introduction

The Lattice Boltzmann Method (LBM) is a powerful tool for simulating fluid dynamics, but it has a significant weakness regarding numerical stability. Standard models, such as the single-relaxation-time LBGK, often crash or produce unphysical "wiggles" (spurious oscillations) when dealing with sharp gradients. These situations are common in high-speed (high Mach number) or low-viscosity (high Reynolds number) flows.

While grid refinement is a common fix, it is computationally expensive and does not always guarantee convergence for nonlinear systems with shocks. This paper investigates how to "regularize" the simulation—essentially adding targeted dissipation to suppress numerical noise without compromising the physical accuracy of the signal.

Methodology

The authors categorize and test several "add-ons" to the standard LBGK model designed to improve stability through controlled dissipation:

  • Entropic LBM (ELBM): This approach enforces the discrete $H$-theorem, ensuring that entropy always increases (or stays constant) during collisions. It uses a variable parameter $\alpha$ to maintain this balance.
  • Flux Limiters & Entropic Filtering: Borrowing ideas from image processing and traditional CFD, these methods detect "noisy" areas (using the Kullback distance or non-equilibrium entropy) and apply local corrections. A standout is the Median Entropy Filter, which replaces local entropy values with a neighborhood median to suppress "salt and pepper" numerical noise while preserving sharp physical features.
  • Multiple Relaxation Times (MRT): Unlike LBGK, which relaxes all components of the fluid at the same rate, MRT allows different physical modes (like shear and bulk viscosity) to relax independently. This provides additional degrees of freedom that can be tuned to improve stability.
  • Aggressive MRT Limiters: The authors propose using MRT as a limiter, sending non-physical modes directly to equilibrium at specific high-gradient sites. This process can be visualized as a "positivity fixation," where the distribution functions are mapped back into a stable domain.
Geometric representation of positivity fixation within the positivity domain.

Experimental Evaluation

The methods were evaluated using established mathematical benchmarks in computational fluid dynamics (CFD):

  1. 1D Shock Tube: A classic test for capturing sharp discontinuities and managing dispersive oscillations.
  2. 2D Shear Decay: Used to validate that the new methods still produce the correct physical viscosity.
  3. 2D Lid-Driven Cavity: A square box with a moving lid, used to test stability at high Reynolds numbers ($Re$) and to locate the first Hopf Bifurcation (the point where steady flow becomes periodic).

Key Insights and Results

One of the most surprising findings is that precisely maintaining the entropy balance (ELBM) does not automatically suppress spurious oscillations in shock waves. This confirms that thermodynamic correctness is not a substitute for proper numerical regularization. Instead, the core philosophy presented is that regularization is "controlled dissipation," applied only where needed (high-gradient regions) and only to non-hydrodynamic modes.

Stability and Accuracy

  • Stability Gains: While standard LBGK failed at higher Reynolds numbers, the "add-on" methods remained stable. For the lid-driven cavity, simulations were successfully run up to $Re = 10,000$ using the Ehrenfest limiter.
  • Bifurcation Accuracy: Using the median filter limiter on a $100 \times 100$ grid, the authors localized the first Hopf bifurcation at $Re \approx 7135$. This is remarkably close to high-resolution literature values ($Re \approx 7402 \pm 4\%$), proving that these limiters do not "smear" important physics.
  • MRT Performance: MRT models produced shear viscosity at almost the same rate as LBGK but offered significantly better stability by contracting the non-equilibrium population space.

Conclusion and Future Work

While these add-ons significantly extend the stable operating range of LBM, some limitations remain. Pointwise filtering can sometimes introduce small-scale local oscillations, suggesting the system might not be perfectly stable in the long-term limit without careful threshold tuning. Furthermore, choosing optimal relaxation rates for MRT remains an open question. Future research will focus on extending these entropic limiters to non-isothermal flows with significant temperature variations.

Summary updated

Abstract

We describe how regularization of lattice Boltzmann methods can be achieved by modifying dissipation. Classes of techniques used to try to improve regularization of LBMs include flux limiters, enforcing the exact correct production of entropy and manipulating non-hydrodynamic modes of the system in relaxation. Each of these techniques corresponds to an additional modification of dissipation compared with the standard LBGK model. Using some standard 1D and 2D benchmarks including the shock tube and lid driven cavity, we explore the effectiveness of these classes of methods.
BibTeX
@inproceedings{brownlee2013add,
  author = {R. A. Brownlee and Jeremy Levesley and David Packwood and Alexander N. Gorban},
  title = {{Add-ons for Lattice Boltzmann Methods: Regularization, Filtering and Limiters}},
  booktitle = {BENTHAM},
  year = {2013},
  doi = {10.2174/9781608057160113030005},
}
Figure