YAC's parallelization scheme requires that each process has to be able to locally determine the owner of a vertex in the YAC-internal decomposition. This operation has produced the same result on all processes. With high compiler optimisation, this requirement can be broken. The following steps should improve overall numerical robustness: - [x] develop arbitrary precision algorithm for determine position of vertex in respect to a great circle plane (use orient3d used by Hongyu as reference Make use of this algorithm in all sphere part operations. - [ ] store a norm vector for each edge and make sure it is bit identical on all processes, use it in clipping - [ ] make sure vertices are bit identical on all processes - [ ] add option to exclude user-defined components from weight computation (might be useful in heterogeneous setups)