“…They generate very rich visual results, reproducing the complexity of running fluids at the price of high computation costs; except for [9] which deals with 2D fluids and [22] that introduces a scheme allowing stability even out of the time step range required by the physical simulation. -Signal processing approaches [19,21,4,13,24], also based on intensive calculations (far less than CFD, however), can be qualified as 'phenomenological' in that they aim at reproducing the effects (spatio-temporal shape or force field) without looking at the causes, oppositely to CFD. The problem is that statistical models tend to lose persistent features (e.g., eddies) and that a good model accounting for static distribution (e.g., for clouds density) doesn't trivially yield a good animated model.…”