An Efficient Linear Octree-Based Grid Toward Magnetic Fluid Stimulation

An Eulerian approach to fluid flow provides an efficient stable paradigm for realistic fluid simulation. However its reliance on a fixed-resolution grid is not ideal for simulations that exhibit both large and small-scale fluid phenomena. A coarse grid can efficiently capture large-scale effects like ocean waves, but lacks the resolution needed for small-scale phenomena like droplets. On the other hand, a fine grid can capture these small-scale effects, but is inefficient for large-scale phenomena. Magnetic fluid, or ferrofluid, illustrates this problem with its very fine detail centered about a magnet and lack of detail elsewhere. Our new fluid simulation technique builds upon previous octree-based methods by simulating on a custom linear octree-based grid structure. A linear octree is stored contiguously in memory rather than as a recursive set of pointers. This use of memory improves the cache coherency issues inherent in previous octree methods. By localizing high-resolution regions we can allow the simulation of small-scale phenomena, while at the same time maintaining efficiency in coarse grid regions. We believe our new simulation technique will provide a framework for simulating ferrofluids which have not been simulated in a physically-based manner from a computer graphics perspective as of yet