Grid-based Finite Elements System for Solving Laplace-Beltrami Equations on 2-Manifolds
MetadataShow full item record
Solving the Poisson equation has numerous important applications. On a Riemannian 2-manifold, the task is most often formulated in terms of finite elements and two challenges commonly arise: discretizing the space of functions and solving the resulting system of equations. In this work, we describe a finite elements system that simultaneously addresses both aspects. The idea is to define a space of functions in 3D and then restrict the 3D functions to the mesh. Unlike traditional approaches, our method is tessellation-independent and has a direct control over system complexity. More importantly, the resulting function space comes with a multi-resolution structure supporting an efficient multigrid solver, and the regularity of the function space can be leveraged in parallelizing/streaming the computation. We evaluate our framework by conducting several experiments. These include a spectral analysis that reveals the embedding-invariant robustness of our discretization, and a benchmark for solver convergence/performance that reveals the competitiveness of our approach against other state-of-the-art methods. We apply our work to several geometry-processing applications. Using curvature flows, we show that we can support efficient surface evolution where the embedding changes with time. Formulating surface filtering as a solution to the screened-Poisson equation, we demonstrate that we can support an anisotropic surface editing system that processes high resolution meshes in real time.