Karman Vortex street simulation

A simple Karman street simulation that uses a D2Q9 LBM CFD solver for real-time web.

A genuine discretized D2Q9 LBM model on a grid measuring 320x180 cells, with boundary interactions handling halfway bounceback logic, which establishes actual no-slip conditions around the cylinder (no shortcuts taken).
High-efficiency density and velocity clamping before the collision step (U_max = 0.2) to make sure there are no particle explosions. The BGK collision relaxes the clamped values back into the physical bounds.
Adjustable values for the f = St(U/D) formula (sliders for flow velocity U, diameter D, kinematic viscocity v) to make sure the LBM field reacts to the Strouhal principle, with the Reynolds number Re provided as well
Optimized for JIT compilers to reduce object allocations on the hot path, uses CDNs/mirrors for global resource coverage

Multi-relaxation time (MRT) is not implemented in favor of the BGK single-relaxation scalar even though it’s unconditionally stable, because the transformation matrix requires more flops on a non-accelerated backend
The Smagorinsky model requires recalculation of the nonequilibrium strain rate tensor S_ij for every cell just to add artificial viscocity, which is unideal for real-time contexts
Zou/He & Boudini interpolated bounceback boundaries are also not implemented because they are better for moving curves while the object in the simulation is strictly a stationary cylinder. Hence a staircasing halfway bounceback model is sufficient.
Overall, more advanced methods used in industrial solvers like ANSYS Fluent aren’t implemented because this one is a good approximation for demo and simulation

Looks pretty cool CFD_Frame

Embedded simulation (view the simulation in full-screen here):