3D Navier–Stokes / GPU Predictor–Corrector

Spire — Gaudí Tower, Manhattan

Wind field collision study · Lower Manhattan, New York

— fps · grid — · — cells

mean |u| — m/s peak — m/s

div residual — step —

ν — jacobi — iters tracers —

Method. A staggered Eulerian grid solves incompressible Navier–Stokes in real time using a predictor–corrector scheme: a semi‑Lagrangian prediction step advects velocity and smoke density backward through the field, then a Jacobi pressure solve corrects the result to be divergence‑free, with vorticity confinement restoring sub‑grid turbulence. The tower cluster is baked into the grid as a solid boundary — wind and fog genuinely deflect, separate and re‑circulate around its spires.

Tracer colour — vorticity / turbulence

laminar moderate turbulent

fog volume — ray-marched density, solar-lit

Drag to orbit · scroll to dolly

Flow field

Wind speed 1.00

Direction 205°

Fog source 1.00

Turbulence 1.40

Solver parameters

Viscosity ν 0.04

NS cycles × 1.00×

Tracer density 100%

ν smooths velocity gradients.
cycles× scales Jacobi pressure iterations.
Higher = better div-free convergence.

Render

Fog volume

Streamlines

Solver grid

Camera