Parallel tetrahedral mesh generation

Complex industrial applications, e.g. in fluid mechanics, require finer and finer discretizations with meshes made of hundreds of millions of tetrahedra. Today's mesh generators are essentially sequential and are not able to produce more than ten millions of tetrahedra per minute, which makes mesh generation the main bottleneck of engineering analysis. This thesis primarily addresses the efficiency of tetrahedral mesh generation. It presents a new parallelization strategy based on the Hilbert curve, with unprecedented performance results. Finite element solvers are also very sensitive to the quality of meshes: even a few badly shaped tetrahedra can make a simulation unreliable. Finding better solutions for generating high quality meshes is the second challenge that this thesis tackles.