Stable simulation of fluid flow with high-Reynolds number using Ehrenfests' steps

R. A. Brownlee, Alexander N. Gorban, Jeremy LevesleyNumer. Algorithms 2007 · OpenReview

Abstract

The simulation of high-Reynolds number fluid flows remains a significant challenge in computational fluid dynamics due to the inherent instabilities of numerical schemes. This paper presents a novel regularization technique for the Lattice Boltzmann Method (LBM) based on Ehrenfests' coarse-graining. By periodically resetting the simulation state to local equilibrium—a process termed an "Ehrenfests' step"—we demonstrate a robust approach to stabilizing simulations of turbulent and high-speed flows. Our method allows for the decoupling of time steps and viscosity, achieving high accuracy with significantly reduced computational requirements compared to standard LBM approaches.

Introduction

Simulating fluid flow at high Reynolds numbers ($Re$), where inertial forces dominate viscous ones, is notoriously difficult. In these regimes, fluids develop sharp gradients and complex turbulent structures. The Lattice Boltzmann Method (LBM) is a popular alternative to traditional Navier-Stokes solvers because it models fluids as collections of particles on a grid. However, LBM suffers from a "stability cliff": as viscosity decreases (increasing $Re$), the simulation often encounters numerical explosions where values diverge to infinity.

Traditional solutions to this instability involve either massive grid refinement, which is computationally expensive, or the introduction of artificial viscosity, which compromises physical accuracy. This work introduces a stabilizer derived from the statistical mechanics concept of Ehrenfests' coarse-graining to bridge the gap between stability and efficiency.

Theoretical Foundation: Ehrenfests' Coarse-Graining

The core of this approach lies in the mathematical demonstration that the Navier-Stokes equations emerge naturally from a cycle of "free-flight" dynamics followed by "equilibration." In this framework, the fluid state is periodically projected onto a manifold of local equilibrium states.

This theoretical link allows us to treat the simulation as a series of Ehrenfests' steps. By resetting the particle distribution to its local equilibrium after a period of motion, we can effectively regularize the simulation without losing the underlying physics of the fluid flow. This provides a flexible foundation for constructing computational models that remain stable even in near-zero viscosity limits.

Methodology

To implement this theory into a practical numerical scheme, we introduce a selective regularization process:

  1. Nonequilibrium Entropy Monitoring: At every grid point, we calculate the "nonequilibrium entropy" ($\Delta S$). This metric quantifies how far the local particle distribution has drifted from its physical equilibrium.
  2. The Ehrenfests' Step: When the entropy deviation exceeds a predefined threshold ($\delta$), the simulation performs an Ehrenfests' step at that specific location, forcing the particles back to their equilibrium state.
  3. Selective Stabilization: To maintain second-order accuracy, the "reset" is not applied globally. Instead, it is applied only to a small, fixed number ($k$) of the most unstable sites at any given time step. This surgical application of the stabilizer prevents numerical crashes while preserving the fine details of the flow.

Experimental Validation

The method was evaluated using two classic computational fluid dynamics benchmarks:

  • 1D Shock Tube: This test assesses the scheme's ability to handle sharp pressure jumps and extremely low viscosity ($\nu \sim 10^{-9}$). The proposed method successfully suppressed the spurious oscillations that typically cause standard LBM models to fail.
  • 2D Flow Past a Square Cylinder: This benchmark focuses on capturing "vortex shedding"—the oscillating wake created behind an object. The results were compared against Okajima’s experimental data, a standard reference for this phenomenon.

Key Results and Discussion

The Ehrenfests' step approach demonstrated remarkable performance in both stability and efficiency:

  • High-Reynolds Stability: The team successfully simulated vortex shedding at a Reynolds number of 20,000, a regime where standard LBM is traditionally highly unstable.
  • Computational Efficiency: For the square cylinder benchmark, a standard LBM simulation would have required approximately $10^7$ lattice sites to maintain stability. Our method achieved comparable results using only $10^5$ sites—a 100-fold reduction in grid requirements.
  • Physical Accuracy: The calculated Strouhal number (representing the frequency of vortex shedding) closely matched experimental physical data across a wide range of Reynolds numbers, confirming that the regularization does not degrade the physical validity of the simulation.

Conclusion

By utilizing entropy as a local diagnostic tool and applying Ehrenfests' steps as a "safety valve," we have developed a robust regularizer for LBM. This approach allows for stable, high-accuracy simulations of high-Reynolds number flows with a fraction of the computational cost of traditional methods. Future work will focus on the practical implementation of this entropy-based monitoring in massive 3D simulations and refining the selection criteria for the stabilization parameters $k$ and $\delta$.

Summary updated

Abstract

The Navier–Stokes equations arise naturally as a result of Ehrenfests’ coarse-graining in phase space after a period of free-flight dynamics. This point of view allows for a very flexible approach to the simulation of fluid flow for high-Reynolds number. We construct regularisers for lattice Boltzmann computational models. These regularisers are based on Ehrenfests’ coarse-graining idea and could be applied to schemes with either entropic or non-entropic quasiequilibria. We give a numerical scheme which gives good results for the standard test cases of the shock tube and the flow past a square cylinder.
BibTeX
@inproceedings{brownlee2007stable,
  author = {R. A. Brownlee and Alexander N. Gorban and Jeremy Levesley},
  title = {{Stable simulation of fluid flow with high-Reynolds number using Ehrenfests' steps}},
  booktitle = {Numer. Algorithms},
  year = {2007},
  url = {https://openreview.net/forum?id=jca-o-e8FK3},
}
Figure