Fast interface tracking via a multiresolution representation of curves and surfaces (Q2388996): Difference between revisions

The author describes an interface tracking algorithm that uses a multiresolution representation of the interface instead of point values. This represention is based on normal meshes, which is an efficient way to describe curves and surfaces. The main result is a proof, backed by numerical experimentation, that, for fixed small enough accuracy, the cost of the method is only \(O(\log{N}/\Delta t)\) or even \(O(1/\Delta t)\) for \(N\) markers and timestep \(\Delta t\). Thus, the cost to propagate the whole curve is of the same order as the cost of propagating just one point, that is, an order better than for front tracking and level sets. The interface is represented by wavelet vectors which correspond to the details of the interface on different scale levels. It is known that for a fixed curve or surface, the size of these vectors decays rapidly as the scale becomes finer. In the dynamic setting, it is shown that the time derivatives of the wavelet vectors decay in a similar manner. This means that the fine scales evolve more slowly than the coarse scales of the interface. Taking shorter timesteps for the coarse scales than for the fine scales greatly reduces computation cost without affecting the overall accuracy. Extensions to higher dimensions and co-dimensions are also considered.