A technique that simulates fluids faster by tracking particles only near the liquid surface instead of throughout the entire volume. Reduces 24 million particles to 1 million and cuts runtime and memory roughly in half, though naive implementations can cause instability.
Dear fellow scholars, this is two minute papers with our endeavors in creating amazingly detailed fluid simulations is often hamstrung by the fact that we need to simulate the motion of tens of millions of particles. Needless to say, this means excruciatingly long computation times and large memory consumption. This piece of work tries to alleviate the problem by confining the usage of particles to a narrow band close to the liquid surface, and thus decimating the number of particles used in the simulation. The rest of the simulation is computed on a very coarse grid where we compute quantities of the fluid, like velocity and pressure, in grid points, and instead of computing them everywhere, we try to guess what is happening between these grid points. The drawback of this is that we may miss a lot of details because of that. And the brilliant part of this new technique is that we only use a cheap, sparse grid where there is not a lot of things happening, and use the expensive particles only near the surface, where there are a lot of details we can capture. Well. The flip term that you see in the video means fluid implicit particle, a popular way of simulating fluids that uses both grids and particles. In this scene, the old method uses 24 million particles, while the new technique uses only 1 million and creates closely matching results. You can see a lot of excess particles in the footage with the classical simulation technique, and the foamish looking version is the proposed new, more efficient algorithm. Creating such a technique is anything but trivial. Unless special measures are taken, the simulation may have robustness issues, which means that there are situations where it does not produce a sensible result. This is demonstrated in a few examples where with the naive version of the technique, a piece of fluid never ever comes to rest or it may exhibit behaviors that are clearly unstable. It also takes approximately half as much time to run the simulation and uses half as much memory, which is a huge relief for visual effects artists. I don't know about you, fellow scholars, but I see a flood of amazing fluid papers coming in the near future and I'm having quite a bit of trouble containing my excitement. Exciting times are ahead indeed. Thanks for watching and for your generous support, and I'll see you next time.
