[hsnyder]

Hey everyone. I gather that some of the people on this forum (including some of the more prolific posters) work with Lattice Boltzmann approaches as well as the more traditional Navier Stokes formulations of CFD. I'd be very curious to hear from those with experience writing code for both approaches, what you perceive the relative strengths and weaknesses of each to be? In what situations would you reach for one over the other?



I'm familiar with some of the most basic differences... namely that Lattice Boltzmann approaches parallelize very well, but irregular boundaries can be a challenge, as can high speeds (Mach number needs to be low). But it seems to me that you could do away with the lattice and adopt a discrete Boltzmann approach based on a finite element method (or similar), and avoid some of the issues with irregular domains (I've seen that done in a few papers). Similarly it seems to me like you could find some way to formulate a Navier Stokes solver that avoided global-to-global communication and hence was just as parallelizable as the Lattice Boltzmann method (example: explicit time integration, discontinuous galerkin spatial discretization). This is mostly speculation on my part, I'd be very interested to hear some opinions from people with actual experience.