Model Reduction of Complex Dynamics

Model Reduction of Complex Dynamics



Alright, for some actual news. This is something that has been hard to achieve in computer graphics--that is believable fluid mechanics like water or smoke. Traditionally, they've been modeled as really large particle systems, which gets to be pretty expensive computationally.

These guys have managed to reduce the amount of computation required to make these computations to simulate fluids, detailed in this paper.

I've only glanced at the paper, but it looks like they were able to frame the problem in such a way that they were able to use dimension reduction techniques to reduce the number of computations they need to do, but have the least noticeable effect. By noticeable, I mean not only to the eye, but also to physics. It also conserves kinetic energy in the simulation.

I don't understand much of the math in there, as it'll take some time to go through it, but it reminds me of lossy compression algorithms and search engines. Not every piece of information is important, or is important in the same way. If you can frame the problem so that you can throw away less important information and still have approximately the same thing. It's kind of novel to think of it as applied to a computational process, rather than a stream of data.