Coreixas DSFD Abstract 2020
Coreixas DSFD Abstract 2020
Coreixas DSFD Abstract 2020
During the past three decades, the lattice Boltzmann method (LBM) was proven to be a
serious alternative to Navier-Stokes-Fourier solvers for both academic and industry groups. For
the former, this is mainly explained by the versatility and implementation simplicity of LBMs
for the simulations of different types of flow (multiphase, multicomponent, reactive, relativistic,
etc) and physical phenomena (electromagnetism, quantum systems, etc). Regarding people from
industry, they grew a strong interest in this numerical tool because of its natural coupling with
Cartesian grid meshes, corresponding mesh refinement techniques, and its inherent advantage for
parallel computations, all of which leading to a drastic reduction of the pre-processing and wall-
clock times of high-fidelity simulations based on realistic geometries.
Nevertheless, the simple BGK-LBM is known to suffer from stability issues in the context of
high-Reynolds number flows and/or for under-resolved conditions. As a remedy, a large number of
collision models have been proposed, most of the time in their own framework, and with various
degrees of success. They usually rely on a particular moment-space-based collision operator (raw,
Hermite, central, central Hermite, cumulant), a regularization procedure (projection based or
recursive), or a dynamic computation of relaxation frequencies (entropic, subgrid scale models,
spatial filtering). Unfortunately, their efficiency in terms of accuracy, stability and CPU time is
rarely studied in a comprehensive manner.
In this context, the present work aims at comparing the most common collision models through
linear stability analyses (LSA) and validation testcases of increasing complexity: vortex convection,
double shear layer, dipole-wall interaction, flow past an airfoil, etc. While LSA help (1) under-
standing the effect of multirelaxation time (MRT) formulations on the stability of LBMs and (2)
optimizing relaxation frequencies accordingly, the numerical validation (based on Palabos software
[1]) further allows us to compare collision models in terms of accuracy and CPU time using both
(optimal) static and dynamic MRT formulations. Here, the focus is put on D2Q9-LBMs [2,3], but
preliminary results on D3Q19 and D3Q27 formulations are also proposed [4]. In the end, this work
provides a first idea as to which collision models are the most appropriate for the simulation of
high-Reynolds number flows in the isothermal and (weakly) compressible regime.
[1] Latt et al., Palabos: Parallel Lattice Boltzmann Solver, 2019, under review.
[2] Coreixas, Wissocq, Chopard and Latt, Impact of collision models on the physical properties and the
stability of lattice Boltzmann methods, 2020, arXiv:2002.05265, under review.
[3] Coreixas et al., Comprehensive comparison of collision models in the lattice Boltzmann framework:
Impact on the stability and accuracy of D2Q9 models, 2020, under preparation.
[4] Coreixas et al., Comprehensive comparison of collision models in the lattice Boltzmann framework:
Focus on D3Q19 and D3Q27 models, 2020, under preparation.